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"
51 #include "bytestream.h"
59 /* the different Cook versions */
60 #define MONO 0x1000001
61 #define STEREO 0x1000002
62 #define JOINT_STEREO 0x1000003
63 #define MC_COOK 0x2000000
65 #define SUBBAND_SIZE 20
66 #define MAX_SUBPACKETS 5
68 typedef struct cook_gains {
73 typedef struct COOKSubpacket {
81 int samples_per_channel;
82 int log2_numvector_size;
83 unsigned int channel_mask;
86 int bits_per_subpacket;
89 int numvector_size; // 1 << log2_numvector_size;
91 float mono_previous_buffer1[1024];
92 float mono_previous_buffer2[1024];
102 typedef struct cook {
104 * The following 5 functions provide the lowlevel arithmetic on
105 * the internal audio buffers.
107 void (*scalar_dequant)(struct cook *q, int index, int quant_index,
108 int *subband_coef_index, int *subband_coef_sign,
111 void (*decouple)(struct cook *q,
115 float *decode_buffer,
116 float *mlt_buffer1, float *mlt_buffer2);
118 void (*imlt_window)(struct cook *q, float *buffer1,
119 cook_gains *gains_ptr, float *previous_buffer);
121 void (*interpolate)(struct cook *q, float *buffer,
122 int gain_index, int gain_index_next);
124 void (*saturate_output)(struct cook *q, float *out);
126 AVCodecContext* avctx;
127 AudioDSPContext adsp;
131 int samples_per_channel;
134 int discarded_packets;
141 VLC envelope_quant_index[13];
142 VLC sqvh[7]; // scalar quantization
144 /* generate tables and related variables */
145 int gain_size_factor;
146 float gain_table[31];
150 uint8_t* decoded_bytes_buffer;
151 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
152 float decode_buffer_1[1024];
153 float decode_buffer_2[1024];
154 float decode_buffer_0[1060]; /* static allocation for joint decode */
156 const float *cplscales[5];
158 COOKSubpacket subpacket[MAX_SUBPACKETS];
161 static float pow2tab[127];
162 static float rootpow2tab[127];
164 /*************** init functions ***************/
166 /* table generator */
167 static av_cold void init_pow2table(void)
169 /* fast way of computing 2^i and 2^(0.5*i) for -63 <= i < 64 */
171 static const float exp2_tab[2] = {1, M_SQRT2};
172 float exp2_val = powf(2, -63);
173 float root_val = powf(2, -32);
174 for (i = -63; i < 64; i++) {
177 pow2tab[63 + i] = exp2_val;
178 rootpow2tab[63 + i] = root_val * exp2_tab[i & 1];
183 /* table generator */
184 static av_cold void init_gain_table(COOKContext *q)
187 q->gain_size_factor = q->samples_per_channel / 8;
188 for (i = 0; i < 31; i++)
189 q->gain_table[i] = pow(pow2tab[i + 48],
190 (1.0 / (double) q->gain_size_factor));
194 static av_cold int init_cook_vlc_tables(COOKContext *q)
199 for (i = 0; i < 13; i++) {
200 result |= ff_init_vlc_from_lengths(&q->envelope_quant_index[i], 9, 24,
201 envelope_quant_index_huffbits[i], 1,
202 envelope_quant_index_huffsyms[i], 1, 1,
205 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
206 for (i = 0; i < 7; i++) {
207 int sym_size = 1 + (i == 3);
208 result |= ff_init_vlc_from_lengths(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
210 cvh_huffsyms[i], sym_size, sym_size,
214 for (i = 0; i < q->num_subpackets; i++) {
215 if (q->subpacket[i].joint_stereo == 1) {
216 result |= ff_init_vlc_from_lengths(&q->subpacket[i].channel_coupling, 6,
217 (1 << q->subpacket[i].js_vlc_bits) - 1,
218 ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1,
219 ccpl_huffsyms[q->subpacket[i].js_vlc_bits - 2], 1, 1,
221 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
225 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
229 static av_cold int init_cook_mlt(COOKContext *q)
232 int mlt_size = q->samples_per_channel;
234 if ((q->mlt_window = av_malloc_array(mlt_size, sizeof(*q->mlt_window))) == 0)
235 return AVERROR(ENOMEM);
237 /* Initialize the MLT window: simple sine window. */
238 ff_sine_window_init(q->mlt_window, mlt_size);
239 for (j = 0; j < mlt_size; j++)
240 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
242 /* Initialize the MDCT. */
243 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
244 av_freep(&q->mlt_window);
247 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
248 av_log2(mlt_size) + 1);
253 static av_cold void init_cplscales_table(COOKContext *q)
256 for (i = 0; i < 5; i++)
257 q->cplscales[i] = cplscales[i];
260 /*************** init functions end ***********/
262 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
263 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
266 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
267 * Why? No idea, some checksum/error detection method maybe.
269 * Out buffer size: extra bytes are needed to cope with
270 * padding/misalignment.
271 * Subpackets passed to the decoder can contain two, consecutive
272 * half-subpackets, of identical but arbitrary size.
273 * 1234 1234 1234 1234 extraA extraB
274 * Case 1: AAAA BBBB 0 0
275 * Case 2: AAAA ABBB BB-- 3 3
276 * Case 3: AAAA AABB BBBB 2 2
277 * Case 4: AAAA AAAB BBBB BB-- 1 5
279 * Nice way to waste CPU cycles.
281 * @param inbuffer pointer to byte array of indata
282 * @param out pointer to byte array of outdata
283 * @param bytes number of bytes
285 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
287 static const uint32_t tab[4] = {
288 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
289 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
294 uint32_t *obuf = (uint32_t *) out;
295 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
296 * I'm too lazy though, should be something like
297 * for (i = 0; i < bitamount / 64; i++)
298 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
299 * Buffer alignment needs to be checked. */
301 off = (intptr_t) inbuffer & 3;
302 buf = (const uint32_t *) (inbuffer - off);
305 for (i = 0; i < bytes / 4; i++)
306 obuf[i] = c ^ buf[i];
311 static av_cold int cook_decode_close(AVCodecContext *avctx)
314 COOKContext *q = avctx->priv_data;
315 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
317 /* Free allocated memory buffers. */
318 av_freep(&q->mlt_window);
319 av_freep(&q->decoded_bytes_buffer);
321 /* Free the transform. */
322 ff_mdct_end(&q->mdct_ctx);
324 /* Free the VLC tables. */
325 for (i = 0; i < 13; i++)
326 ff_free_vlc(&q->envelope_quant_index[i]);
327 for (i = 0; i < 7; i++)
328 ff_free_vlc(&q->sqvh[i]);
329 for (i = 0; i < q->num_subpackets; i++)
330 ff_free_vlc(&q->subpacket[i].channel_coupling);
332 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
338 * Fill the gain array for the timedomain quantization.
340 * @param gb pointer to the GetBitContext
341 * @param gaininfo array[9] of gain indexes
343 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
347 n = get_unary(gb, 0, get_bits_left(gb)); // amount of elements*2 to update
351 int index = get_bits(gb, 3);
352 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
355 gaininfo[i++] = gain;
362 * Create the quant index table needed for the envelope.
364 * @param q pointer to the COOKContext
365 * @param quant_index_table pointer to the array
367 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
368 int *quant_index_table)
372 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
374 for (i = 1; i < p->total_subbands; i++) {
376 if (i >= p->js_subband_start * 2) {
377 vlc_index -= p->js_subband_start;
384 vlc_index = 13; // the VLC tables >13 are identical to No. 13
386 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
387 q->envelope_quant_index[vlc_index - 1].bits, 2);
388 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
389 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
390 av_log(q->avctx, AV_LOG_ERROR,
391 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
392 quant_index_table[i], i);
393 return AVERROR_INVALIDDATA;
401 * Calculate the category and category_index vector.
403 * @param q pointer to the COOKContext
404 * @param quant_index_table pointer to the array
405 * @param category pointer to the category array
406 * @param category_index pointer to the category_index array
408 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
409 int *category, int *category_index)
411 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
412 int exp_index2[102] = { 0 };
413 int exp_index1[102] = { 0 };
415 int tmp_categorize_array[128 * 2] = { 0 };
416 int tmp_categorize_array1_idx = p->numvector_size;
417 int tmp_categorize_array2_idx = p->numvector_size;
419 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
421 if (bits_left > q->samples_per_channel)
422 bits_left = q->samples_per_channel +
423 ((bits_left - q->samples_per_channel) * 5) / 8;
428 for (i = 32; i > 0; i = i / 2) {
431 for (j = p->total_subbands; j > 0; j--) {
432 exp_idx = av_clip_uintp2((i - quant_index_table[index] + bias) / 2, 3);
434 num_bits += expbits_tab[exp_idx];
436 if (num_bits >= bits_left - 32)
440 /* Calculate total number of bits. */
442 for (i = 0; i < p->total_subbands; i++) {
443 exp_idx = av_clip_uintp2((bias - quant_index_table[i]) / 2, 3);
444 num_bits += expbits_tab[exp_idx];
445 exp_index1[i] = exp_idx;
446 exp_index2[i] = exp_idx;
448 tmpbias1 = tmpbias2 = num_bits;
450 for (j = 1; j < p->numvector_size; j++) {
451 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
454 for (i = 0; i < p->total_subbands; i++) {
455 if (exp_index1[i] < 7) {
456 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
465 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
466 tmpbias1 -= expbits_tab[exp_index1[index]] -
467 expbits_tab[exp_index1[index] + 1];
472 for (i = 0; i < p->total_subbands; i++) {
473 if (exp_index2[i] > 0) {
474 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
483 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
484 tmpbias2 -= expbits_tab[exp_index2[index]] -
485 expbits_tab[exp_index2[index] - 1];
490 for (i = 0; i < p->total_subbands; i++)
491 category[i] = exp_index2[i];
493 for (i = 0; i < p->numvector_size - 1; i++)
494 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
499 * Expand the category vector.
501 * @param q pointer to the COOKContext
502 * @param category pointer to the category array
503 * @param category_index pointer to the category_index array
505 static inline void expand_category(COOKContext *q, int *category,
509 for (i = 0; i < q->num_vectors; i++)
511 int idx = category_index[i];
512 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
518 * The real requantization of the mltcoefs
520 * @param q pointer to the COOKContext
522 * @param quant_index quantisation index
523 * @param subband_coef_index array of indexes to quant_centroid_tab
524 * @param subband_coef_sign signs of coefficients
525 * @param mlt_p pointer into the mlt buffer
527 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
528 int *subband_coef_index, int *subband_coef_sign,
534 for (i = 0; i < SUBBAND_SIZE; i++) {
535 if (subband_coef_index[i]) {
536 f1 = quant_centroid_tab[index][subband_coef_index[i]];
537 if (subband_coef_sign[i])
540 /* noise coding if subband_coef_index[i] == 0 */
541 f1 = dither_tab[index];
542 if (av_lfg_get(&q->random_state) < 0x80000000)
545 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
549 * Unpack the subband_coef_index and subband_coef_sign vectors.
551 * @param q pointer to the COOKContext
552 * @param category pointer to the category array
553 * @param subband_coef_index array of indexes to quant_centroid_tab
554 * @param subband_coef_sign signs of coefficients
556 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
557 int *subband_coef_index, int *subband_coef_sign)
560 int vlc, vd, tmp, result;
562 vd = vd_tab[category];
564 for (i = 0; i < vpr_tab[category]; i++) {
565 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
566 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
570 for (j = vd - 1; j >= 0; j--) {
571 tmp = (vlc * invradix_tab[category]) / 0x100000;
572 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
575 for (j = 0; j < vd; j++) {
576 if (subband_coef_index[i * vd + j]) {
577 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
578 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
581 subband_coef_sign[i * vd + j] = 0;
584 subband_coef_sign[i * vd + j] = 0;
593 * Fill the mlt_buffer with mlt coefficients.
595 * @param q pointer to the COOKContext
596 * @param category pointer to the category array
597 * @param quant_index_table pointer to the array
598 * @param mlt_buffer pointer to mlt coefficients
600 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
601 int *quant_index_table, float *mlt_buffer)
603 /* A zero in this table means that the subband coefficient is
604 random noise coded. */
605 int subband_coef_index[SUBBAND_SIZE];
606 /* A zero in this table means that the subband coefficient is a
607 positive multiplicator. */
608 int subband_coef_sign[SUBBAND_SIZE];
612 for (band = 0; band < p->total_subbands; band++) {
613 index = category[band];
614 if (category[band] < 7) {
615 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
617 for (j = 0; j < p->total_subbands; j++)
618 category[band + j] = 7;
622 memset(subband_coef_index, 0, sizeof(subband_coef_index));
623 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
625 q->scalar_dequant(q, index, quant_index_table[band],
626 subband_coef_index, subband_coef_sign,
627 &mlt_buffer[band * SUBBAND_SIZE]);
630 /* FIXME: should this be removed, or moved into loop above? */
631 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
636 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
638 int category_index[128] = { 0 };
639 int category[128] = { 0 };
640 int quant_index_table[102];
643 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
645 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
646 categorize(q, p, quant_index_table, category, category_index);
647 expand_category(q, category, category_index);
648 for (i=0; i<p->total_subbands; i++) {
650 return AVERROR_INVALIDDATA;
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[15 + (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
751 static int decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
754 int vlc = get_bits1(&q->gb);
755 int start = cplband[p->js_subband_start];
756 int end = cplband[p->subbands - 1];
757 int length = end - start + 1;
763 for (i = 0; i < length; i++)
764 decouple_tab[start + i] = get_vlc2(&q->gb,
765 p->channel_coupling.table,
766 p->channel_coupling.bits, 3);
768 for (i = 0; i < length; i++) {
769 int v = get_bits(&q->gb, p->js_vlc_bits);
770 if (v == (1<<p->js_vlc_bits)-1) {
771 av_log(q->avctx, AV_LOG_ERROR, "decouple value too large\n");
772 return AVERROR_INVALIDDATA;
774 decouple_tab[start + i] = v;
780 * function decouples a pair of signals from a single signal via multiplication.
782 * @param q pointer to the COOKContext
783 * @param subband index of the current subband
784 * @param f1 multiplier for channel 1 extraction
785 * @param f2 multiplier for channel 2 extraction
786 * @param decode_buffer input buffer
787 * @param mlt_buffer1 pointer to left channel mlt coefficients
788 * @param mlt_buffer2 pointer to right channel mlt coefficients
790 static void decouple_float(COOKContext *q,
794 float *decode_buffer,
795 float *mlt_buffer1, float *mlt_buffer2)
798 for (j = 0; j < SUBBAND_SIZE; j++) {
799 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
800 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
801 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
806 * function for decoding joint stereo data
808 * @param q pointer to the COOKContext
809 * @param mlt_buffer1 pointer to left channel mlt coefficients
810 * @param mlt_buffer2 pointer to right channel mlt coefficients
812 static int joint_decode(COOKContext *q, COOKSubpacket *p,
813 float *mlt_buffer_left, float *mlt_buffer_right)
816 int decouple_tab[SUBBAND_SIZE] = { 0 };
817 float *decode_buffer = q->decode_buffer_0;
820 const float *cplscale;
822 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
824 /* Make sure the buffers are zeroed out. */
825 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
826 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
827 if ((res = decouple_info(q, p, decouple_tab)) < 0)
829 if ((res = mono_decode(q, p, decode_buffer)) < 0)
831 /* The two channels are stored interleaved in decode_buffer. */
832 for (i = 0; i < p->js_subband_start; i++) {
833 for (j = 0; j < SUBBAND_SIZE; j++) {
834 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
835 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
839 /* When we reach js_subband_start (the higher frequencies)
840 the coefficients are stored in a coupling scheme. */
841 idx = (1 << p->js_vlc_bits) - 1;
842 for (i = p->js_subband_start; i < p->subbands; i++) {
843 cpl_tmp = cplband[i];
844 idx -= decouple_tab[cpl_tmp];
845 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
846 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
848 q->decouple(q, p, i, f1, f2, decode_buffer,
849 mlt_buffer_left, mlt_buffer_right);
850 idx = (1 << p->js_vlc_bits) - 1;
857 * First part of subpacket decoding:
858 * decode raw stream bytes and read gain info.
860 * @param q pointer to the COOKContext
861 * @param inbuffer pointer to raw stream data
862 * @param gains_ptr array of current/prev gain pointers
864 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
865 const uint8_t *inbuffer,
866 cook_gains *gains_ptr)
870 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
871 p->bits_per_subpacket / 8);
872 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
873 p->bits_per_subpacket);
874 decode_gain_info(&q->gb, gains_ptr->now);
876 /* Swap current and previous gains */
877 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
881 * Saturate the output signal and interleave.
883 * @param q pointer to the COOKContext
884 * @param out pointer to the output vector
886 static void saturate_output_float(COOKContext *q, float *out)
888 q->adsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
889 FFALIGN(q->samples_per_channel, 8), -1.0f, 1.0f);
894 * Final part of subpacket decoding:
895 * Apply modulated lapped transform, gain compensation,
896 * clip and convert to integer.
898 * @param q pointer to the COOKContext
899 * @param decode_buffer pointer to the mlt coefficients
900 * @param gains_ptr array of current/prev gain pointers
901 * @param previous_buffer pointer to the previous buffer to be used for overlapping
902 * @param out pointer to the output buffer
904 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
905 cook_gains *gains_ptr, float *previous_buffer,
908 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
910 q->saturate_output(q, out);
915 * Cook subpacket decoding. This function returns one decoded subpacket,
916 * usually 1024 samples per channel.
918 * @param q pointer to the COOKContext
919 * @param inbuffer pointer to the inbuffer
920 * @param outbuffer pointer to the outbuffer
922 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
923 const uint8_t *inbuffer, float **outbuffer)
925 int sub_packet_size = p->size;
928 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
929 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
931 if (p->joint_stereo) {
932 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
935 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
938 if (p->num_channels == 2) {
939 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
940 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
945 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
946 p->mono_previous_buffer1,
947 outbuffer ? outbuffer[p->ch_idx] : NULL);
949 if (p->num_channels == 2) {
951 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
952 p->mono_previous_buffer2,
953 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
955 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
956 p->mono_previous_buffer2,
957 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
964 static int cook_decode_frame(AVCodecContext *avctx, void *data,
965 int *got_frame_ptr, AVPacket *avpkt)
967 AVFrame *frame = data;
968 const uint8_t *buf = avpkt->data;
969 int buf_size = avpkt->size;
970 COOKContext *q = avctx->priv_data;
971 float **samples = NULL;
976 if (buf_size < avctx->block_align)
979 /* get output buffer */
980 if (q->discarded_packets >= 2) {
981 frame->nb_samples = q->samples_per_channel;
982 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
984 samples = (float **)frame->extended_data;
987 /* estimate subpacket sizes */
988 q->subpacket[0].size = avctx->block_align;
990 for (i = 1; i < q->num_subpackets; i++) {
991 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
992 q->subpacket[0].size -= q->subpacket[i].size + 1;
993 if (q->subpacket[0].size < 0) {
994 av_log(avctx, AV_LOG_DEBUG,
995 "frame subpacket size total > avctx->block_align!\n");
996 return AVERROR_INVALIDDATA;
1000 /* decode supbackets */
1001 for (i = 0; i < q->num_subpackets; i++) {
1002 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
1003 q->subpacket[i].bits_per_subpdiv;
1004 q->subpacket[i].ch_idx = chidx;
1005 av_log(avctx, AV_LOG_DEBUG,
1006 "subpacket[%i] size %i js %i %i block_align %i\n",
1007 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1008 avctx->block_align);
1010 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1012 offset += q->subpacket[i].size;
1013 chidx += q->subpacket[i].num_channels;
1014 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1015 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1018 /* Discard the first two frames: no valid audio. */
1019 if (q->discarded_packets < 2) {
1020 q->discarded_packets++;
1022 return avctx->block_align;
1027 return avctx->block_align;
1030 static void dump_cook_context(COOKContext *q)
1033 #define PRINT(a, b) ff_dlog(q->avctx, " %s = %d\n", a, b);
1034 ff_dlog(q->avctx, "COOKextradata\n");
1035 ff_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1036 if (q->subpacket[0].cookversion > STEREO) {
1037 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1038 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1040 ff_dlog(q->avctx, "COOKContext\n");
1041 PRINT("nb_channels", q->avctx->channels);
1042 PRINT("bit_rate", (int)q->avctx->bit_rate);
1043 PRINT("sample_rate", q->avctx->sample_rate);
1044 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1045 PRINT("subbands", q->subpacket[0].subbands);
1046 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1047 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1048 PRINT("numvector_size", q->subpacket[0].numvector_size);
1049 PRINT("total_subbands", q->subpacket[0].total_subbands);
1053 * Cook initialization
1055 * @param avctx pointer to the AVCodecContext
1057 static av_cold int cook_decode_init(AVCodecContext *avctx)
1059 COOKContext *q = avctx->priv_data;
1062 unsigned int channel_mask = 0;
1063 int samples_per_frame = 0;
1067 /* Take care of the codec specific extradata. */
1068 if (avctx->extradata_size < 8) {
1069 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1070 return AVERROR_INVALIDDATA;
1072 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1074 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1076 /* Take data from the AVCodecContext (RM container). */
1077 if (!avctx->channels) {
1078 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1079 return AVERROR_INVALIDDATA;
1082 if (avctx->block_align >= INT_MAX / 8)
1083 return AVERROR(EINVAL);
1085 /* Initialize RNG. */
1086 av_lfg_init(&q->random_state, 0);
1088 ff_audiodsp_init(&q->adsp);
1090 while (bytestream2_get_bytes_left(&gb)) {
1091 if (s >= FFMIN(MAX_SUBPACKETS, avctx->block_align)) {
1092 avpriv_request_sample(avctx, "subpackets > %d", FFMIN(MAX_SUBPACKETS, avctx->block_align));
1093 return AVERROR_PATCHWELCOME;
1095 /* 8 for mono, 16 for stereo, ? for multichannel
1096 Swap to right endianness so we don't need to care later on. */
1097 q->subpacket[s].cookversion = bytestream2_get_be32(&gb);
1098 samples_per_frame = bytestream2_get_be16(&gb);
1099 q->subpacket[s].subbands = bytestream2_get_be16(&gb);
1100 bytestream2_get_be32(&gb); // Unknown unused
1101 q->subpacket[s].js_subband_start = bytestream2_get_be16(&gb);
1102 if (q->subpacket[s].js_subband_start >= 51) {
1103 av_log(avctx, AV_LOG_ERROR, "js_subband_start %d is too large\n", q->subpacket[s].js_subband_start);
1104 return AVERROR_INVALIDDATA;
1106 q->subpacket[s].js_vlc_bits = bytestream2_get_be16(&gb);
1108 /* Initialize extradata related variables. */
1109 q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1110 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1112 /* Initialize default data states. */
1113 q->subpacket[s].log2_numvector_size = 5;
1114 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1115 q->subpacket[s].num_channels = 1;
1117 /* Initialize version-dependent variables */
1119 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1120 q->subpacket[s].cookversion);
1121 q->subpacket[s].joint_stereo = 0;
1122 switch (q->subpacket[s].cookversion) {
1124 if (avctx->channels != 1) {
1125 avpriv_request_sample(avctx, "Container channels != 1");
1126 return AVERROR_PATCHWELCOME;
1128 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1131 if (avctx->channels != 1) {
1132 q->subpacket[s].bits_per_subpdiv = 1;
1133 q->subpacket[s].num_channels = 2;
1135 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1138 if (avctx->channels != 2) {
1139 avpriv_request_sample(avctx, "Container channels != 2");
1140 return AVERROR_PATCHWELCOME;
1142 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1143 if (avctx->extradata_size >= 16) {
1144 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1145 q->subpacket[s].js_subband_start;
1146 q->subpacket[s].joint_stereo = 1;
1147 q->subpacket[s].num_channels = 2;
1149 if (q->subpacket[s].samples_per_channel > 256) {
1150 q->subpacket[s].log2_numvector_size = 6;
1152 if (q->subpacket[s].samples_per_channel > 512) {
1153 q->subpacket[s].log2_numvector_size = 7;
1157 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1158 channel_mask |= q->subpacket[s].channel_mask = bytestream2_get_be32(&gb);
1160 if (av_get_channel_layout_nb_channels(q->subpacket[s].channel_mask) > 1) {
1161 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1162 q->subpacket[s].js_subband_start;
1163 q->subpacket[s].joint_stereo = 1;
1164 q->subpacket[s].num_channels = 2;
1165 q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1167 if (q->subpacket[s].samples_per_channel > 256) {
1168 q->subpacket[s].log2_numvector_size = 6;
1170 if (q->subpacket[s].samples_per_channel > 512) {
1171 q->subpacket[s].log2_numvector_size = 7;
1174 q->subpacket[s].samples_per_channel = samples_per_frame;
1178 avpriv_request_sample(avctx, "Cook version %d",
1179 q->subpacket[s].cookversion);
1180 return AVERROR_PATCHWELCOME;
1183 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1184 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1185 return AVERROR_INVALIDDATA;
1187 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1190 /* Initialize variable relations */
1191 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1193 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1194 if (q->subpacket[s].total_subbands > 53) {
1195 avpriv_request_sample(avctx, "total_subbands > 53");
1196 return AVERROR_PATCHWELCOME;
1199 if ((q->subpacket[s].js_vlc_bits > 6) ||
1200 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1201 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1202 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1203 return AVERROR_INVALIDDATA;
1206 if (q->subpacket[s].subbands > 50) {
1207 avpriv_request_sample(avctx, "subbands > 50");
1208 return AVERROR_PATCHWELCOME;
1210 if (q->subpacket[s].subbands == 0) {
1211 avpriv_request_sample(avctx, "subbands = 0");
1212 return AVERROR_PATCHWELCOME;
1214 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1215 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1216 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1217 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1219 if (q->num_subpackets + q->subpacket[s].num_channels > q->avctx->channels) {
1220 av_log(avctx, AV_LOG_ERROR, "Too many subpackets %d for channels %d\n", q->num_subpackets, q->avctx->channels);
1221 return AVERROR_INVALIDDATA;
1224 q->num_subpackets++;
1228 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1229 if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1230 q->samples_per_channel != 1024) {
1231 avpriv_request_sample(avctx, "samples_per_channel = %d",
1232 q->samples_per_channel);
1233 return AVERROR_PATCHWELCOME;
1236 /* Generate tables */
1239 init_cplscales_table(q);
1241 if ((ret = init_cook_vlc_tables(q)))
1244 /* Pad the databuffer with:
1245 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1246 AV_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 + AV_INPUT_BUFFER_PADDING_SIZE);
1251 if (!q->decoded_bytes_buffer)
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 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1269 avctx->channel_layout = channel_mask;
1271 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1274 dump_cook_context(q);
1279 AVCodec ff_cook_decoder = {
1281 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1282 .type = AVMEDIA_TYPE_AUDIO,
1283 .id = AV_CODEC_ID_COOK,
1284 .priv_data_size = sizeof(COOKContext),
1285 .init = cook_decode_init,
1286 .close = cook_decode_close,
1287 .decode = cook_decode_frame,
1288 .capabilities = AV_CODEC_CAP_DR1,
1289 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1290 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1291 AV_SAMPLE_FMT_NONE },