blob: 3f7953c6db4b3ab281db150149e0360760b7f961
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 | int 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 | |
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 | |
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 | |
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 | |
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 = (val << bits) | get_bits(gb, bits); |
224 | return val; |
225 | } |
226 | |
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 | ctx->filter_bits = (ctx->filter_params - 2) >> 6; |
238 | ctx->filter_length = ctx->filter_params - (ctx->filter_bits << 6) - 1; |
239 | |
240 | if (ctx->filter_params == FILTER_RAW) { |
241 | for (i = 0; i < length; i++) |
242 | dst[i] = get_bits(gb, bits); |
243 | ctx->bias[ch] = 0; |
244 | return 0; |
245 | } |
246 | |
247 | ctx->bias[ch] = get_vlc2(gb, set->bias.table, 9, 2); |
248 | ctx->bias[ch] = extend_code(gb, ctx->bias[ch], 127, 4); |
249 | |
250 | if (ctx->filter_params == FILTER_NONE) { |
251 | memset(dst, 0, sizeof(*dst) * length); |
252 | return 0; |
253 | } |
254 | |
255 | if (ctx->filter_params > 1) { |
256 | int cmode = 0, coeff = 0; |
257 | VLC *vlc = set->filter_coeffs[ctx->filter_bits] + 5; |
258 | |
259 | add_bits = ctx->filter_bits; |
260 | |
261 | for (i = 0; i < ctx->filter_length; i++) { |
262 | t = get_vlc2(gb, vlc[cmode].table, vlc[cmode].bits, 2); |
263 | t = extend_code(gb, t, 21, add_bits); |
264 | if (!cmode) |
265 | coeff -= 12 << add_bits; |
266 | coeff = t - coeff; |
267 | ctx->filter[i] = coeff; |
268 | |
269 | cmode = coeff >> add_bits; |
270 | if (cmode < 0) { |
271 | cmode = -1 - av_log2(-cmode); |
272 | if (cmode < -5) |
273 | cmode = -5; |
274 | } else if (cmode > 0) { |
275 | cmode = 1 + av_log2(cmode); |
276 | if (cmode > 5) |
277 | cmode = 5; |
278 | } |
279 | } |
280 | } |
281 | |
282 | code_params = get_vlc2(gb, set->coding_mode.table, set->coding_mode.bits, 2); |
283 | if (code_params >= 15) { |
284 | add_bits = av_clip((code_params / 5 - 3) / 2, 0, 10); |
285 | if (add_bits > 9 && (code_params % 5) != 2) |
286 | add_bits--; |
287 | range = 10; |
288 | range2 = 21; |
289 | code_vlc = set->long_codes + code_params - 15; |
290 | } else { |
291 | add_bits = 0; |
292 | range = 6; |
293 | range2 = 13; |
294 | code_vlc = set->short_codes + code_params; |
295 | } |
296 | |
297 | for (i = 0; i < length; i += 2) { |
298 | int code1, code2; |
299 | |
300 | t = get_vlc2(gb, code_vlc->table, code_vlc->bits, 2); |
301 | code1 = t / range2; |
302 | code2 = t % range2; |
303 | dst[i] = extend_code(gb, code1, range, 0) << add_bits; |
304 | dst[i + 1] = extend_code(gb, code2, range, 0) << add_bits; |
305 | if (add_bits) { |
306 | dst[i] |= get_bits(gb, add_bits); |
307 | dst[i + 1] |= get_bits(gb, add_bits); |
308 | } |
309 | } |
310 | |
311 | return 0; |
312 | } |
313 | |
314 | static void apply_lpc(RALFContext *ctx, int ch, int length, int bits) |
315 | { |
316 | int i, j, acc; |
317 | int *audio = ctx->channel_data[ch]; |
318 | int bias = 1 << (ctx->filter_bits - 1); |
319 | int max_clip = (1 << bits) - 1, min_clip = -max_clip - 1; |
320 | |
321 | for (i = 1; i < length; i++) { |
322 | int flen = FFMIN(ctx->filter_length, i); |
323 | |
324 | acc = 0; |
325 | for (j = 0; j < flen; j++) |
326 | acc += ctx->filter[j] * audio[i - j - 1]; |
327 | if (acc < 0) { |
328 | acc = (acc + bias - 1) >> ctx->filter_bits; |
329 | acc = FFMAX(acc, min_clip); |
330 | } else { |
331 | acc = (acc + bias) >> ctx->filter_bits; |
332 | acc = FFMIN(acc, max_clip); |
333 | } |
334 | audio[i] += acc; |
335 | } |
336 | } |
337 | |
338 | static int decode_block(AVCodecContext *avctx, GetBitContext *gb, |
339 | int16_t *dst0, int16_t *dst1) |
340 | { |
341 | RALFContext *ctx = avctx->priv_data; |
342 | int len, ch, ret; |
343 | int dmode, mode[2], bits[2]; |
344 | int *ch0, *ch1; |
345 | int i, t, t2; |
346 | |
347 | len = 12 - get_unary(gb, 0, 6); |
348 | |
349 | if (len <= 7) len ^= 1; // codes for length = 6 and 7 are swapped |
350 | len = 1 << len; |
351 | |
352 | if (ctx->sample_offset + len > ctx->max_frame_size) { |
353 | av_log(avctx, AV_LOG_ERROR, |
354 | "Decoder's stomach is crying, it ate too many samples\n"); |
355 | return AVERROR_INVALIDDATA; |
356 | } |
357 | |
358 | if (avctx->channels > 1) |
359 | dmode = get_bits(gb, 2) + 1; |
360 | else |
361 | dmode = 0; |
362 | |
363 | mode[0] = (dmode == 4) ? 1 : 0; |
364 | mode[1] = (dmode >= 2) ? 2 : 0; |
365 | bits[0] = 16; |
366 | bits[1] = (mode[1] == 2) ? 17 : 16; |
367 | |
368 | for (ch = 0; ch < avctx->channels; ch++) { |
369 | if ((ret = decode_channel(ctx, gb, ch, len, mode[ch], bits[ch])) < 0) |
370 | return ret; |
371 | if (ctx->filter_params > 1 && ctx->filter_params != FILTER_RAW) { |
372 | ctx->filter_bits += 3; |
373 | apply_lpc(ctx, ch, len, bits[ch]); |
374 | } |
375 | if (get_bits_left(gb) < 0) |
376 | return AVERROR_INVALIDDATA; |
377 | } |
378 | ch0 = ctx->channel_data[0]; |
379 | ch1 = ctx->channel_data[1]; |
380 | switch (dmode) { |
381 | case 0: |
382 | for (i = 0; i < len; i++) |
383 | dst0[i] = ch0[i] + ctx->bias[0]; |
384 | break; |
385 | case 1: |
386 | for (i = 0; i < len; i++) { |
387 | dst0[i] = ch0[i] + ctx->bias[0]; |
388 | dst1[i] = ch1[i] + ctx->bias[1]; |
389 | } |
390 | break; |
391 | case 2: |
392 | for (i = 0; i < len; i++) { |
393 | ch0[i] += ctx->bias[0]; |
394 | dst0[i] = ch0[i]; |
395 | dst1[i] = ch0[i] - (ch1[i] + ctx->bias[1]); |
396 | } |
397 | break; |
398 | case 3: |
399 | for (i = 0; i < len; i++) { |
400 | t = ch0[i] + ctx->bias[0]; |
401 | t2 = ch1[i] + ctx->bias[1]; |
402 | dst0[i] = t + t2; |
403 | dst1[i] = t; |
404 | } |
405 | break; |
406 | case 4: |
407 | for (i = 0; i < len; i++) { |
408 | t = ch1[i] + ctx->bias[1]; |
409 | t2 = ((ch0[i] + ctx->bias[0]) << 1) | (t & 1); |
410 | dst0[i] = (t2 + t) / 2; |
411 | dst1[i] = (t2 - t) / 2; |
412 | } |
413 | break; |
414 | } |
415 | |
416 | ctx->sample_offset += len; |
417 | |
418 | return 0; |
419 | } |
420 | |
421 | static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, |
422 | AVPacket *avpkt) |
423 | { |
424 | RALFContext *ctx = avctx->priv_data; |
425 | AVFrame *frame = data; |
426 | int16_t *samples0; |
427 | int16_t *samples1; |
428 | int ret; |
429 | GetBitContext gb; |
430 | int table_size, table_bytes, i; |
431 | const uint8_t *src, *block_pointer; |
432 | int src_size; |
433 | int bytes_left; |
434 | |
435 | if (ctx->has_pkt) { |
436 | ctx->has_pkt = 0; |
437 | table_bytes = (AV_RB16(avpkt->data) + 7) >> 3; |
438 | if (table_bytes + 3 > avpkt->size || avpkt->size > RALF_MAX_PKT_SIZE) { |
439 | av_log(avctx, AV_LOG_ERROR, "Wrong packet's breath smells of wrong data!\n"); |
440 | return AVERROR_INVALIDDATA; |
441 | } |
442 | if (memcmp(ctx->pkt, avpkt->data, 2 + table_bytes)) { |
443 | av_log(avctx, AV_LOG_ERROR, "Wrong packet tails are wrong!\n"); |
444 | return AVERROR_INVALIDDATA; |
445 | } |
446 | |
447 | src = ctx->pkt; |
448 | src_size = RALF_MAX_PKT_SIZE + avpkt->size; |
449 | memcpy(ctx->pkt + RALF_MAX_PKT_SIZE, avpkt->data + 2 + table_bytes, |
450 | avpkt->size - 2 - table_bytes); |
451 | } else { |
452 | if (avpkt->size == RALF_MAX_PKT_SIZE) { |
453 | memcpy(ctx->pkt, avpkt->data, avpkt->size); |
454 | ctx->has_pkt = 1; |
455 | *got_frame_ptr = 0; |
456 | |
457 | return avpkt->size; |
458 | } |
459 | src = avpkt->data; |
460 | src_size = avpkt->size; |
461 | } |
462 | |
463 | frame->nb_samples = ctx->max_frame_size; |
464 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
465 | return ret; |
466 | samples0 = (int16_t *)frame->data[0]; |
467 | samples1 = (int16_t *)frame->data[1]; |
468 | |
469 | if (src_size < 5) { |
470 | av_log(avctx, AV_LOG_ERROR, "too short packets are too short!\n"); |
471 | return AVERROR_INVALIDDATA; |
472 | } |
473 | table_size = AV_RB16(src); |
474 | table_bytes = (table_size + 7) >> 3; |
475 | if (src_size < table_bytes + 3) { |
476 | av_log(avctx, AV_LOG_ERROR, "short packets are short!\n"); |
477 | return AVERROR_INVALIDDATA; |
478 | } |
479 | init_get_bits(&gb, src + 2, table_size); |
480 | ctx->num_blocks = 0; |
481 | while (get_bits_left(&gb) > 0) { |
482 | ctx->block_size[ctx->num_blocks] = get_bits(&gb, 13 + avctx->channels); |
483 | if (get_bits1(&gb)) { |
484 | ctx->block_pts[ctx->num_blocks] = get_bits(&gb, 9); |
485 | } else { |
486 | ctx->block_pts[ctx->num_blocks] = 0; |
487 | } |
488 | ctx->num_blocks++; |
489 | } |
490 | |
491 | block_pointer = src + table_bytes + 2; |
492 | bytes_left = src_size - table_bytes - 2; |
493 | ctx->sample_offset = 0; |
494 | for (i = 0; i < ctx->num_blocks; i++) { |
495 | if (bytes_left < ctx->block_size[i]) { |
496 | av_log(avctx, AV_LOG_ERROR, "I'm pedaling backwards\n"); |
497 | break; |
498 | } |
499 | init_get_bits(&gb, block_pointer, ctx->block_size[i] * 8); |
500 | if (decode_block(avctx, &gb, samples0 + ctx->sample_offset, |
501 | samples1 + ctx->sample_offset) < 0) { |
502 | av_log(avctx, AV_LOG_ERROR, "Sir, I got carsick in your office. Not decoding the rest of packet.\n"); |
503 | break; |
504 | } |
505 | block_pointer += ctx->block_size[i]; |
506 | bytes_left -= ctx->block_size[i]; |
507 | } |
508 | |
509 | frame->nb_samples = ctx->sample_offset; |
510 | *got_frame_ptr = ctx->sample_offset > 0; |
511 | |
512 | return avpkt->size; |
513 | } |
514 | |
515 | static void decode_flush(AVCodecContext *avctx) |
516 | { |
517 | RALFContext *ctx = avctx->priv_data; |
518 | |
519 | ctx->has_pkt = 0; |
520 | } |
521 | |
522 | |
523 | AVCodec ff_ralf_decoder = { |
524 | .name = "ralf", |
525 | .long_name = NULL_IF_CONFIG_SMALL("RealAudio Lossless"), |
526 | .type = AVMEDIA_TYPE_AUDIO, |
527 | .id = AV_CODEC_ID_RALF, |
528 | .priv_data_size = sizeof(RALFContext), |
529 | .init = decode_init, |
530 | .close = decode_close, |
531 | .decode = decode_frame, |
532 | .flush = decode_flush, |
533 | .capabilities = AV_CODEC_CAP_DR1, |
534 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P, |
535 | AV_SAMPLE_FMT_NONE }, |
536 | }; |
537 |