1 /*
2 * RealAudio Lossless decoder
3 *
4 * Copyright (c) 2012 Konstantin Shishkov
5 *
6 * This file is part of FFmpeg.
7 *
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.
12 *
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.
17 *
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
21 */
22
23 /**
24 * @file
25 * This is a decoder for Real Audio Lossless format.
26 * Dedicated to the mastermind behind it, Ralph Wiggum.
27 */
28
29 #include "libavutil/attributes.h"
30 #include "libavutil/channel_layout.h"
31 #include "avcodec.h"
32 #include "get_bits.h"
33 #include "golomb.h"
34 #include "internal.h"
35 #include "unary.h"
36 #include "ralfdata.h"
37
38 #define FILTER_NONE 0
39 #define FILTER_RAW 642
40
41 typedef struct VLCSet {
42 VLC filter_params;
43 VLC bias;
44 VLC coding_mode;
45 VLC filter_coeffs[10][11];
46 VLC short_codes[15];
47 VLC long_codes[125];
48 } VLCSet;
49
50 #define RALF_MAX_PKT_SIZE 8192
51
52 typedef struct RALFContext {
53 int version;
54 int max_frame_size;
55 VLCSet sets[3];
56 int32_t channel_data[2][4096];
57
58 int filter_params; ///< combined filter parameters for the current channel data
59 int filter_length; ///< length of the filter for the current channel data
60 int filter_bits; ///< filter precision for the current channel data
61 int32_t filter[64];
62
63 unsigned bias[2]; ///< a constant value added to channel data after filtering
64
65 int num_blocks; ///< number of blocks inside the frame
66 int sample_offset;
67 int block_size[1 << 12]; ///< size of the blocks
68 int block_pts[1 << 12]; ///< block start time (in milliseconds)
69
70 uint8_t pkt[16384];
71 int has_pkt;
72 } RALFContext;
73
74 #define MAX_ELEMS 644 // no RALF table uses more than that
75
init_ralf_vlc(VLC * vlc,const uint8_t * data,int elems)76 static av_cold int init_ralf_vlc(VLC *vlc, const uint8_t *data, int elems)
77 {
78 uint8_t lens[MAX_ELEMS];
79 uint16_t codes[MAX_ELEMS];
80 int counts[17], prefixes[18];
81 int i, cur_len;
82 int max_bits = 0;
83 int nb = 0;
84
85 for (i = 0; i <= 16; i++)
86 counts[i] = 0;
87 for (i = 0; i < elems; i++) {
88 cur_len = (nb ? *data & 0xF : *data >> 4) + 1;
89 counts[cur_len]++;
90 max_bits = FFMAX(max_bits, cur_len);
91 lens[i] = cur_len;
92 data += nb;
93 nb ^= 1;
94 }
95 prefixes[1] = 0;
96 for (i = 1; i <= 16; i++)
97 prefixes[i + 1] = (prefixes[i] + counts[i]) << 1;
98
99 for (i = 0; i < elems; i++)
100 codes[i] = prefixes[lens[i]]++;
101
102 return ff_init_vlc_sparse(vlc, FFMIN(max_bits, 9), elems,
103 lens, 1, 1, codes, 2, 2, NULL, 0, 0, 0);
104 }
105
decode_close(AVCodecContext * avctx)106 static av_cold int decode_close(AVCodecContext *avctx)
107 {
108 RALFContext *ctx = avctx->priv_data;
109 int i, j, k;
110
111 for (i = 0; i < 3; i++) {
112 ff_free_vlc(&ctx->sets[i].filter_params);
113 ff_free_vlc(&ctx->sets[i].bias);
114 ff_free_vlc(&ctx->sets[i].coding_mode);
115 for (j = 0; j < 10; j++)
116 for (k = 0; k < 11; k++)
117 ff_free_vlc(&ctx->sets[i].filter_coeffs[j][k]);
118 for (j = 0; j < 15; j++)
119 ff_free_vlc(&ctx->sets[i].short_codes[j]);
120 for (j = 0; j < 125; j++)
121 ff_free_vlc(&ctx->sets[i].long_codes[j]);
122 }
123
124 return 0;
125 }
126
decode_init(AVCodecContext * avctx)127 static av_cold int decode_init(AVCodecContext *avctx)
128 {
129 RALFContext *ctx = avctx->priv_data;
130 int i, j, k;
131 int ret;
132
133 if (avctx->extradata_size < 24 || memcmp(avctx->extradata, "LSD:", 4)) {
134 av_log(avctx, AV_LOG_ERROR, "Extradata is not groovy, dude\n");
135 return AVERROR_INVALIDDATA;
136 }
137
138 ctx->version = AV_RB16(avctx->extradata + 4);
139 if (ctx->version != 0x103) {
140 avpriv_request_sample(avctx, "Unknown version %X", ctx->version);
141 return AVERROR_PATCHWELCOME;
142 }
143
144 avctx->channels = AV_RB16(avctx->extradata + 8);
145 avctx->sample_rate = AV_RB32(avctx->extradata + 12);
146 if (avctx->channels < 1 || avctx->channels > 2
147 || avctx->sample_rate < 8000 || avctx->sample_rate > 96000) {
148 av_log(avctx, AV_LOG_ERROR, "Invalid coding parameters %d Hz %d ch\n",
149 avctx->sample_rate, avctx->channels);
150 return AVERROR_INVALIDDATA;
151 }
152 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
153 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO
154 : AV_CH_LAYOUT_MONO;
155
156 ctx->max_frame_size = AV_RB32(avctx->extradata + 16);
157 if (ctx->max_frame_size > (1 << 20) || !ctx->max_frame_size) {
158 av_log(avctx, AV_LOG_ERROR, "invalid frame size %d\n",
159 ctx->max_frame_size);
160 }
161 ctx->max_frame_size = FFMAX(ctx->max_frame_size, avctx->sample_rate);
162
163 for (i = 0; i < 3; i++) {
164 ret = init_ralf_vlc(&ctx->sets[i].filter_params, filter_param_def[i],
165 FILTERPARAM_ELEMENTS);
166 if (ret < 0) {
167 decode_close(avctx);
168 return ret;
169 }
170 ret = init_ralf_vlc(&ctx->sets[i].bias, bias_def[i], BIAS_ELEMENTS);
171 if (ret < 0) {
172 decode_close(avctx);
173 return ret;
174 }
175 ret = init_ralf_vlc(&ctx->sets[i].coding_mode, coding_mode_def[i],
176 CODING_MODE_ELEMENTS);
177 if (ret < 0) {
178 decode_close(avctx);
179 return ret;
180 }
181 for (j = 0; j < 10; j++) {
182 for (k = 0; k < 11; k++) {
183 ret = init_ralf_vlc(&ctx->sets[i].filter_coeffs[j][k],
184 filter_coeffs_def[i][j][k],
185 FILTER_COEFFS_ELEMENTS);
186 if (ret < 0) {
187 decode_close(avctx);
188 return ret;
189 }
190 }
191 }
192 for (j = 0; j < 15; j++) {
193 ret = init_ralf_vlc(&ctx->sets[i].short_codes[j],
194 short_codes_def[i][j], SHORT_CODES_ELEMENTS);
195 if (ret < 0) {
196 decode_close(avctx);
197 return ret;
198 }
199 }
200 for (j = 0; j < 125; j++) {
201 ret = init_ralf_vlc(&ctx->sets[i].long_codes[j],
202 long_codes_def[i][j], LONG_CODES_ELEMENTS);
203 if (ret < 0) {
204 decode_close(avctx);
205 return ret;
206 }
207 }
208 }
209
210 return 0;
211 }
212
extend_code(GetBitContext * gb,int val,int range,int bits)213 static inline int extend_code(GetBitContext *gb, int val, int range, int bits)
214 {
215 if (val == 0) {
216 val = -range - get_ue_golomb(gb);
217 } else if (val == range * 2) {
218 val = range + get_ue_golomb(gb);
219 } else {
220 val -= range;
221 }
222 if (bits)
223 val = ((unsigned)val << bits) | get_bits(gb, bits);
224 return val;
225 }
226
decode_channel(RALFContext * ctx,GetBitContext * gb,int ch,int length,int mode,int bits)227 static int decode_channel(RALFContext *ctx, GetBitContext *gb, int ch,
228 int length, int mode, int bits)
229 {
230 int i, t;
231 int code_params;
232 VLCSet *set = ctx->sets + mode;
233 VLC *code_vlc; int range, range2, add_bits;
234 int *dst = ctx->channel_data[ch];
235
236 ctx->filter_params = get_vlc2(gb, set->filter_params.table, 9, 2);
237 if (ctx->filter_params > 1) {
238 ctx->filter_bits = (ctx->filter_params - 2) >> 6;
239 ctx->filter_length = ctx->filter_params - (ctx->filter_bits << 6) - 1;
240 }
241
242 if (ctx->filter_params == FILTER_RAW) {
243 for (i = 0; i < length; i++)
244 dst[i] = get_bits(gb, bits);
245 ctx->bias[ch] = 0;
246 return 0;
247 }
248
249 ctx->bias[ch] = get_vlc2(gb, set->bias.table, 9, 2);
250 ctx->bias[ch] = extend_code(gb, ctx->bias[ch], 127, 4);
251
252 if (ctx->filter_params == FILTER_NONE) {
253 memset(dst, 0, sizeof(*dst) * length);
254 return 0;
255 }
256
257 if (ctx->filter_params > 1) {
258 int cmode = 0, coeff = 0;
259 VLC *vlc = set->filter_coeffs[ctx->filter_bits] + 5;
260
261 add_bits = ctx->filter_bits;
262
263 for (i = 0; i < ctx->filter_length; i++) {
264 t = get_vlc2(gb, vlc[cmode].table, vlc[cmode].bits, 2);
265 t = extend_code(gb, t, 21, add_bits);
266 if (!cmode)
267 coeff -= 12U << add_bits;
268 coeff = (unsigned)t - coeff;
269 ctx->filter[i] = coeff;
270
271 cmode = coeff >> add_bits;
272 if (cmode < 0) {
273 cmode = -1 - av_log2(-cmode);
274 if (cmode < -5)
275 cmode = -5;
276 } else if (cmode > 0) {
277 cmode = 1 + av_log2(cmode);
278 if (cmode > 5)
279 cmode = 5;
280 }
281 }
282 }
283
284 code_params = get_vlc2(gb, set->coding_mode.table, set->coding_mode.bits, 2);
285 if (code_params >= 15) {
286 add_bits = av_clip((code_params / 5 - 3) / 2, 0, 10);
287 if (add_bits > 9 && (code_params % 5) != 2)
288 add_bits--;
289 range = 10;
290 range2 = 21;
291 code_vlc = set->long_codes + (code_params - 15);
292 } else {
293 add_bits = 0;
294 range = 6;
295 range2 = 13;
296 code_vlc = set->short_codes + code_params;
297 }
298
299 for (i = 0; i < length; i += 2) {
300 int code1, code2;
301
302 t = get_vlc2(gb, code_vlc->table, code_vlc->bits, 2);
303 code1 = t / range2;
304 code2 = t % range2;
305 dst[i] = extend_code(gb, code1, range, 0) * (1U << add_bits);
306 dst[i + 1] = extend_code(gb, code2, range, 0) * (1U << add_bits);
307 if (add_bits) {
308 dst[i] |= get_bits(gb, add_bits);
309 dst[i + 1] |= get_bits(gb, add_bits);
310 }
311 }
312
313 return 0;
314 }
315
apply_lpc(RALFContext * ctx,int ch,int length,int bits)316 static void apply_lpc(RALFContext *ctx, int ch, int length, int bits)
317 {
318 int i, j, acc;
319 int *audio = ctx->channel_data[ch];
320 int bias = 1 << (ctx->filter_bits - 1);
321 int max_clip = (1 << bits) - 1, min_clip = -max_clip - 1;
322
323 for (i = 1; i < length; i++) {
324 int flen = FFMIN(ctx->filter_length, i);
325
326 acc = 0;
327 for (j = 0; j < flen; j++)
328 acc += (unsigned)ctx->filter[j] * audio[i - j - 1];
329 if (acc < 0) {
330 acc = (acc + bias - 1) >> ctx->filter_bits;
331 acc = FFMAX(acc, min_clip);
332 } else {
333 acc = ((unsigned)acc + bias) >> ctx->filter_bits;
334 acc = FFMIN(acc, max_clip);
335 }
336 audio[i] += acc;
337 }
338 }
339
decode_block(AVCodecContext * avctx,GetBitContext * gb,int16_t * dst0,int16_t * dst1)340 static int decode_block(AVCodecContext *avctx, GetBitContext *gb,
341 int16_t *dst0, int16_t *dst1)
342 {
343 RALFContext *ctx = avctx->priv_data;
344 int len, ch, ret;
345 int dmode, mode[2], bits[2];
346 int *ch0, *ch1;
347 int i;
348 unsigned int t, t2;
349
350 len = 12 - get_unary(gb, 0, 6);
351
352 if (len <= 7) len ^= 1; // codes for length = 6 and 7 are swapped
353 len = 1 << len;
354
355 if (ctx->sample_offset + len > ctx->max_frame_size) {
356 av_log(avctx, AV_LOG_ERROR,
357 "Decoder's stomach is crying, it ate too many samples\n");
358 return AVERROR_INVALIDDATA;
359 }
360
361 if (avctx->channels > 1)
362 dmode = get_bits(gb, 2) + 1;
363 else
364 dmode = 0;
365
366 mode[0] = (dmode == 4) ? 1 : 0;
367 mode[1] = (dmode >= 2) ? 2 : 0;
368 bits[0] = 16;
369 bits[1] = (mode[1] == 2) ? 17 : 16;
370
371 for (ch = 0; ch < avctx->channels; ch++) {
372 if ((ret = decode_channel(ctx, gb, ch, len, mode[ch], bits[ch])) < 0)
373 return ret;
374 if (ctx->filter_params > 1 && ctx->filter_params != FILTER_RAW) {
375 ctx->filter_bits += 3;
376 apply_lpc(ctx, ch, len, bits[ch]);
377 }
378 if (get_bits_left(gb) < 0)
379 return AVERROR_INVALIDDATA;
380 }
381 ch0 = ctx->channel_data[0];
382 ch1 = ctx->channel_data[1];
383 switch (dmode) {
384 case 0:
385 for (i = 0; i < len; i++)
386 dst0[i] = ch0[i] + ctx->bias[0];
387 break;
388 case 1:
389 for (i = 0; i < len; i++) {
390 dst0[i] = ch0[i] + ctx->bias[0];
391 dst1[i] = ch1[i] + ctx->bias[1];
392 }
393 break;
394 case 2:
395 for (i = 0; i < len; i++) {
396 ch0[i] += ctx->bias[0];
397 dst0[i] = ch0[i];
398 dst1[i] = ch0[i] - (ch1[i] + ctx->bias[1]);
399 }
400 break;
401 case 3:
402 for (i = 0; i < len; i++) {
403 t = ch0[i] + ctx->bias[0];
404 t2 = ch1[i] + ctx->bias[1];
405 dst0[i] = t + t2;
406 dst1[i] = t;
407 }
408 break;
409 case 4:
410 for (i = 0; i < len; i++) {
411 t = ch1[i] + ctx->bias[1];
412 t2 = ((ch0[i] + ctx->bias[0]) * 2) | (t & 1);
413 dst0[i] = (int)(t2 + t) / 2;
414 dst1[i] = (int)(t2 - t) / 2;
415 }
416 break;
417 }
418
419 ctx->sample_offset += len;
420
421 return 0;
422 }
423
decode_frame(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)424 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
425 AVPacket *avpkt)
426 {
427 RALFContext *ctx = avctx->priv_data;
428 AVFrame *frame = data;
429 int16_t *samples0;
430 int16_t *samples1;
431 int ret;
432 GetBitContext gb;
433 int table_size, table_bytes, i;
434 const uint8_t *src, *block_pointer;
435 int src_size;
436 int bytes_left;
437
438 if (ctx->has_pkt) {
439 ctx->has_pkt = 0;
440 table_bytes = (AV_RB16(avpkt->data) + 7) >> 3;
441 if (table_bytes + 3 > avpkt->size || avpkt->size > RALF_MAX_PKT_SIZE) {
442 av_log(avctx, AV_LOG_ERROR, "Wrong packet's breath smells of wrong data!\n");
443 return AVERROR_INVALIDDATA;
444 }
445 if (memcmp(ctx->pkt, avpkt->data, 2 + table_bytes)) {
446 av_log(avctx, AV_LOG_ERROR, "Wrong packet tails are wrong!\n");
447 return AVERROR_INVALIDDATA;
448 }
449
450 src = ctx->pkt;
451 src_size = RALF_MAX_PKT_SIZE + avpkt->size;
452 memcpy(ctx->pkt + RALF_MAX_PKT_SIZE, avpkt->data + 2 + table_bytes,
453 avpkt->size - 2 - table_bytes);
454 } else {
455 if (avpkt->size == RALF_MAX_PKT_SIZE) {
456 memcpy(ctx->pkt, avpkt->data, avpkt->size);
457 ctx->has_pkt = 1;
458 *got_frame_ptr = 0;
459
460 return avpkt->size;
461 }
462 src = avpkt->data;
463 src_size = avpkt->size;
464 }
465
466 frame->nb_samples = ctx->max_frame_size;
467 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
468 return ret;
469 samples0 = (int16_t *)frame->data[0];
470 samples1 = (int16_t *)frame->data[1];
471
472 if (src_size < 5) {
473 av_log(avctx, AV_LOG_ERROR, "too short packets are too short!\n");
474 return AVERROR_INVALIDDATA;
475 }
476 table_size = AV_RB16(src);
477 table_bytes = (table_size + 7) >> 3;
478 if (src_size < table_bytes + 3) {
479 av_log(avctx, AV_LOG_ERROR, "short packets are short!\n");
480 return AVERROR_INVALIDDATA;
481 }
482 init_get_bits(&gb, src + 2, table_size);
483 ctx->num_blocks = 0;
484 while (get_bits_left(&gb) > 0) {
485 if (ctx->num_blocks >= FF_ARRAY_ELEMS(ctx->block_size))
486 return AVERROR_INVALIDDATA;
487 ctx->block_size[ctx->num_blocks] = get_bits(&gb, 13 + avctx->channels);
488 if (get_bits1(&gb)) {
489 ctx->block_pts[ctx->num_blocks] = get_bits(&gb, 9);
490 } else {
491 ctx->block_pts[ctx->num_blocks] = 0;
492 }
493 ctx->num_blocks++;
494 }
495
496 block_pointer = src + table_bytes + 2;
497 bytes_left = src_size - table_bytes - 2;
498 ctx->sample_offset = 0;
499 for (i = 0; i < ctx->num_blocks; i++) {
500 if (bytes_left < ctx->block_size[i]) {
501 av_log(avctx, AV_LOG_ERROR, "I'm pedaling backwards\n");
502 break;
503 }
504 init_get_bits(&gb, block_pointer, ctx->block_size[i] * 8);
505 if (decode_block(avctx, &gb, samples0 + ctx->sample_offset,
506 samples1 + ctx->sample_offset) < 0) {
507 av_log(avctx, AV_LOG_ERROR, "Sir, I got carsick in your office. Not decoding the rest of packet.\n");
508 break;
509 }
510 block_pointer += ctx->block_size[i];
511 bytes_left -= ctx->block_size[i];
512 }
513
514 frame->nb_samples = ctx->sample_offset;
515 *got_frame_ptr = ctx->sample_offset > 0;
516
517 return avpkt->size;
518 }
519
decode_flush(AVCodecContext * avctx)520 static void decode_flush(AVCodecContext *avctx)
521 {
522 RALFContext *ctx = avctx->priv_data;
523
524 ctx->has_pkt = 0;
525 }
526
527
528 AVCodec ff_ralf_decoder = {
529 .name = "ralf",
530 .long_name = NULL_IF_CONFIG_SMALL("RealAudio Lossless"),
531 .type = AVMEDIA_TYPE_AUDIO,
532 .id = AV_CODEC_ID_RALF,
533 .priv_data_size = sizeof(RALFContext),
534 .init = decode_init,
535 .close = decode_close,
536 .decode = decode_frame,
537 .flush = decode_flush,
538 .capabilities = AV_CODEC_CAP_CHANNEL_CONF |
539 AV_CODEC_CAP_DR1,
540 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
541 AV_SAMPLE_FMT_NONE },
542 };
543