2 * RealAudio Lossless decoder
4 * Copyright (c) 2012 Konstantin Shishkov
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * This is a decoder for Real Audio Lossless format.
26 * Dedicated to the mastermind behind it, Ralph Wiggum.
29 #include "libavutil/channel_layout.h"
38 #define FILTER_RAW 642
40 typedef struct VLCSet {
44 VLC filter_coeffs[10][11];
49 #define RALF_MAX_PKT_SIZE 8192
51 typedef struct RALFContext {
55 int32_t channel_data[2][4096];
57 int filter_params; ///< combined filter parameters for the current channel data
58 int filter_length; ///< length of the filter for the current channel data
59 int filter_bits; ///< filter precision for the current channel data
62 int bias[2]; ///< a constant value added to channel data after filtering
64 int num_blocks; ///< number of blocks inside the frame
66 int block_size[1 << 12]; ///< size of the blocks
67 int block_pts[1 << 12]; ///< block start time (in milliseconds)
73 #define MAX_ELEMS 644 // no RALF table uses more than that
75 static int init_ralf_vlc(VLC *vlc, const uint8_t *data, int elems)
77 uint8_t lens[MAX_ELEMS];
78 uint16_t codes[MAX_ELEMS];
79 int counts[17], prefixes[18];
84 for (i = 0; i <= 16; i++)
86 for (i = 0; i < elems; i++) {
87 cur_len = (nb ? *data & 0xF : *data >> 4) + 1;
89 max_bits = FFMAX(max_bits, cur_len);
95 for (i = 1; i <= 16; i++)
96 prefixes[i + 1] = (prefixes[i] + counts[i]) << 1;
98 for (i = 0; i < elems; i++)
99 codes[i] = prefixes[lens[i]]++;
101 return ff_init_vlc_sparse(vlc, FFMIN(max_bits, 9), elems,
102 lens, 1, 1, codes, 2, 2, NULL, 0, 0, 0);
105 static av_cold int decode_close(AVCodecContext *avctx)
107 RALFContext *ctx = avctx->priv_data;
110 for (i = 0; i < 3; i++) {
111 ff_free_vlc(&ctx->sets[i].filter_params);
112 ff_free_vlc(&ctx->sets[i].bias);
113 ff_free_vlc(&ctx->sets[i].coding_mode);
114 for (j = 0; j < 10; j++)
115 for (k = 0; k < 11; k++)
116 ff_free_vlc(&ctx->sets[i].filter_coeffs[j][k]);
117 for (j = 0; j < 15; j++)
118 ff_free_vlc(&ctx->sets[i].short_codes[j]);
119 for (j = 0; j < 125; j++)
120 ff_free_vlc(&ctx->sets[i].long_codes[j]);
126 static av_cold int decode_init(AVCodecContext *avctx)
128 RALFContext *ctx = avctx->priv_data;
132 if (avctx->extradata_size < 24 || memcmp(avctx->extradata, "LSD:", 4)) {
133 av_log(avctx, AV_LOG_ERROR, "Extradata is not groovy, dude\n");
134 return AVERROR_INVALIDDATA;
137 ctx->version = AV_RB16(avctx->extradata + 4);
138 if (ctx->version != 0x103) {
139 avpriv_request_sample(avctx, "Unknown version %X", ctx->version);
140 return AVERROR_PATCHWELCOME;
143 avctx->channels = AV_RB16(avctx->extradata + 8);
144 avctx->sample_rate = AV_RB32(avctx->extradata + 12);
145 if (avctx->channels < 1 || avctx->channels > 2
146 || avctx->sample_rate < 8000 || avctx->sample_rate > 96000) {
147 av_log(avctx, AV_LOG_ERROR, "Invalid coding parameters %d Hz %d ch\n",
148 avctx->sample_rate, avctx->channels);
149 return AVERROR_INVALIDDATA;
151 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
152 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO
155 ctx->max_frame_size = AV_RB32(avctx->extradata + 16);
156 if (ctx->max_frame_size > (1 << 20) || !ctx->max_frame_size) {
157 av_log(avctx, AV_LOG_ERROR, "invalid frame size %d\n",
158 ctx->max_frame_size);
160 ctx->max_frame_size = FFMAX(ctx->max_frame_size, avctx->sample_rate);
162 for (i = 0; i < 3; i++) {
163 ret = init_ralf_vlc(&ctx->sets[i].filter_params, filter_param_def[i],
164 FILTERPARAM_ELEMENTS);
169 ret = init_ralf_vlc(&ctx->sets[i].bias, bias_def[i], BIAS_ELEMENTS);
174 ret = init_ralf_vlc(&ctx->sets[i].coding_mode, coding_mode_def[i],
175 CODING_MODE_ELEMENTS);
180 for (j = 0; j < 10; j++) {
181 for (k = 0; k < 11; k++) {
182 ret = init_ralf_vlc(&ctx->sets[i].filter_coeffs[j][k],
183 filter_coeffs_def[i][j][k],
184 FILTER_COEFFS_ELEMENTS);
191 for (j = 0; j < 15; j++) {
192 ret = init_ralf_vlc(&ctx->sets[i].short_codes[j],
193 short_codes_def[i][j], SHORT_CODES_ELEMENTS);
199 for (j = 0; j < 125; j++) {
200 ret = init_ralf_vlc(&ctx->sets[i].long_codes[j],
201 long_codes_def[i][j], LONG_CODES_ELEMENTS);
212 static inline int extend_code(GetBitContext *gb, int val, int range, int bits)
215 val = -range - get_ue_golomb(gb);
216 } else if (val == range * 2) {
217 val = range + get_ue_golomb(gb);
222 val = (val << bits) | get_bits(gb, bits);
226 static int decode_channel(RALFContext *ctx, GetBitContext *gb, int ch,
227 int length, int mode, int bits)
231 VLCSet *set = ctx->sets + mode;
232 VLC *code_vlc; int range, range2, add_bits;
233 int *dst = ctx->channel_data[ch];
235 ctx->filter_params = get_vlc2(gb, set->filter_params.table, 9, 2);
236 ctx->filter_bits = (ctx->filter_params - 2) >> 6;
237 ctx->filter_length = ctx->filter_params - (ctx->filter_bits << 6) - 1;
239 if (ctx->filter_params == FILTER_RAW) {
240 for (i = 0; i < length; i++)
241 dst[i] = get_bits(gb, bits);
246 ctx->bias[ch] = get_vlc2(gb, set->bias.table, 9, 2);
247 ctx->bias[ch] = extend_code(gb, ctx->bias[ch], 127, 4);
249 if (ctx->filter_params == FILTER_NONE) {
250 memset(dst, 0, sizeof(*dst) * length);
254 if (ctx->filter_params > 1) {
255 int cmode = 0, coeff = 0;
256 VLC *vlc = set->filter_coeffs[ctx->filter_bits] + 5;
258 add_bits = ctx->filter_bits;
260 for (i = 0; i < ctx->filter_length; i++) {
261 t = get_vlc2(gb, vlc[cmode].table, vlc[cmode].bits, 2);
262 t = extend_code(gb, t, 21, add_bits);
264 coeff -= 12 << add_bits;
266 ctx->filter[i] = coeff;
268 cmode = coeff >> add_bits;
270 cmode = -1 - av_log2(-cmode);
273 } else if (cmode > 0) {
274 cmode = 1 + av_log2(cmode);
281 code_params = get_vlc2(gb, set->coding_mode.table, set->coding_mode.bits, 2);
282 if (code_params >= 15) {
283 add_bits = av_clip((code_params / 5 - 3) / 2, 0, 10);
284 if (add_bits > 9 && (code_params % 5) != 2)
288 code_vlc = set->long_codes + code_params - 15;
293 code_vlc = set->short_codes + code_params;
296 for (i = 0; i < length; i += 2) {
299 t = get_vlc2(gb, code_vlc->table, code_vlc->bits, 2);
302 dst[i] = extend_code(gb, code1, range, 0) << add_bits;
303 dst[i + 1] = extend_code(gb, code2, range, 0) << add_bits;
305 dst[i] |= get_bits(gb, add_bits);
306 dst[i + 1] |= get_bits(gb, add_bits);
313 static void apply_lpc(RALFContext *ctx, int ch, int length, int bits)
316 int *audio = ctx->channel_data[ch];
317 int bias = 1 << (ctx->filter_bits - 1);
318 int max_clip = (1 << bits) - 1, min_clip = -max_clip - 1;
320 for (i = 1; i < length; i++) {
321 int flen = FFMIN(ctx->filter_length, i);
324 for (j = 0; j < flen; j++)
325 acc += ctx->filter[j] * audio[i - j - 1];
327 acc = (acc + bias - 1) >> ctx->filter_bits;
328 acc = FFMAX(acc, min_clip);
330 acc = (acc + bias) >> ctx->filter_bits;
331 acc = FFMIN(acc, max_clip);
337 static int decode_block(AVCodecContext *avctx, GetBitContext *gb,
338 int16_t *dst0, int16_t *dst1)
340 RALFContext *ctx = avctx->priv_data;
342 int dmode, mode[2], bits[2];
346 len = 12 - get_unary(gb, 0, 6);
348 if (len <= 7) len ^= 1; // codes for length = 6 and 7 are swapped
351 if (ctx->sample_offset + len > ctx->max_frame_size) {
352 av_log(avctx, AV_LOG_ERROR,
353 "Decoder's stomach is crying, it ate too many samples\n");
354 return AVERROR_INVALIDDATA;
357 if (avctx->channels > 1)
358 dmode = get_bits(gb, 2) + 1;
362 mode[0] = (dmode == 4) ? 1 : 0;
363 mode[1] = (dmode >= 2) ? 2 : 0;
365 bits[1] = (mode[1] == 2) ? 17 : 16;
367 for (ch = 0; ch < avctx->channels; ch++) {
368 if ((ret = decode_channel(ctx, gb, ch, len, mode[ch], bits[ch])) < 0)
370 if (ctx->filter_params > 1 && ctx->filter_params != FILTER_RAW) {
371 ctx->filter_bits += 3;
372 apply_lpc(ctx, ch, len, bits[ch]);
374 if (get_bits_left(gb) < 0)
375 return AVERROR_INVALIDDATA;
377 ch0 = ctx->channel_data[0];
378 ch1 = ctx->channel_data[1];
381 for (i = 0; i < len; i++)
382 dst0[i] = ch0[i] + ctx->bias[0];
385 for (i = 0; i < len; i++) {
386 dst0[i] = ch0[i] + ctx->bias[0];
387 dst1[i] = ch1[i] + ctx->bias[1];
391 for (i = 0; i < len; i++) {
392 ch0[i] += ctx->bias[0];
394 dst1[i] = ch0[i] - (ch1[i] + ctx->bias[1]);
398 for (i = 0; i < len; i++) {
399 t = ch0[i] + ctx->bias[0];
400 t2 = ch1[i] + ctx->bias[1];
406 for (i = 0; i < len; i++) {
407 t = ch1[i] + ctx->bias[1];
408 t2 = ((ch0[i] + ctx->bias[0]) << 1) | (t & 1);
409 dst0[i] = (t2 + t) / 2;
410 dst1[i] = (t2 - t) / 2;
415 ctx->sample_offset += len;
420 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
423 RALFContext *ctx = avctx->priv_data;
424 AVFrame *frame = data;
429 int table_size, table_bytes, i;
430 const uint8_t *src, *block_pointer;
436 table_bytes = (AV_RB16(avpkt->data) + 7) >> 3;
437 if (table_bytes + 3 > avpkt->size || avpkt->size > RALF_MAX_PKT_SIZE) {
438 av_log(avctx, AV_LOG_ERROR, "Wrong packet's breath smells of wrong data!\n");
439 return AVERROR_INVALIDDATA;
441 if (memcmp(ctx->pkt, avpkt->data, 2 + table_bytes)) {
442 av_log(avctx, AV_LOG_ERROR, "Wrong packet tails are wrong!\n");
443 return AVERROR_INVALIDDATA;
447 src_size = RALF_MAX_PKT_SIZE + avpkt->size;
448 memcpy(ctx->pkt + RALF_MAX_PKT_SIZE, avpkt->data + 2 + table_bytes,
449 avpkt->size - 2 - table_bytes);
451 if (avpkt->size == RALF_MAX_PKT_SIZE) {
452 memcpy(ctx->pkt, avpkt->data, avpkt->size);
459 src_size = avpkt->size;
462 frame->nb_samples = ctx->max_frame_size;
463 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
465 samples0 = (int16_t *)frame->data[0];
466 samples1 = (int16_t *)frame->data[1];
469 av_log(avctx, AV_LOG_ERROR, "too short packets are too short!\n");
470 return AVERROR_INVALIDDATA;
472 table_size = AV_RB16(src);
473 table_bytes = (table_size + 7) >> 3;
474 if (src_size < table_bytes + 3) {
475 av_log(avctx, AV_LOG_ERROR, "short packets are short!\n");
476 return AVERROR_INVALIDDATA;
478 init_get_bits(&gb, src + 2, table_size);
480 while (get_bits_left(&gb) > 0) {
481 ctx->block_size[ctx->num_blocks] = get_bits(&gb, 15);
482 if (get_bits1(&gb)) {
483 ctx->block_pts[ctx->num_blocks] = get_bits(&gb, 9);
485 ctx->block_pts[ctx->num_blocks] = 0;
490 block_pointer = src + table_bytes + 2;
491 bytes_left = src_size - table_bytes - 2;
492 ctx->sample_offset = 0;
493 for (i = 0; i < ctx->num_blocks; i++) {
494 if (bytes_left < ctx->block_size[i]) {
495 av_log(avctx, AV_LOG_ERROR, "I'm pedaling backwards\n");
498 init_get_bits(&gb, block_pointer, ctx->block_size[i] * 8);
499 if (decode_block(avctx, &gb, samples0 + ctx->sample_offset,
500 samples1 + ctx->sample_offset) < 0) {
501 av_log(avctx, AV_LOG_ERROR, "Sir, I got carsick in your office. Not decoding the rest of packet.\n");
504 block_pointer += ctx->block_size[i];
505 bytes_left -= ctx->block_size[i];
508 frame->nb_samples = ctx->sample_offset;
509 *got_frame_ptr = ctx->sample_offset > 0;
514 static void decode_flush(AVCodecContext *avctx)
516 RALFContext *ctx = avctx->priv_data;
522 AVCodec ff_ralf_decoder = {
524 .type = AVMEDIA_TYPE_AUDIO,
525 .id = AV_CODEC_ID_RALF,
526 .priv_data_size = sizeof(RALFContext),
528 .close = decode_close,
529 .decode = decode_frame,
530 .flush = decode_flush,
531 .capabilities = CODEC_CAP_DR1,
532 .long_name = NULL_IF_CONFIG_SMALL("RealAudio Lossless"),
533 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
534 AV_SAMPLE_FMT_NONE },