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/attributes.h"
30 #include "libavutil/channel_layout.h"
33 #include "bitstream.h"
41 #define FILTER_RAW 642
43 typedef struct VLCSet {
47 VLC filter_coeffs[10][11];
52 #define RALF_MAX_PKT_SIZE 8192
54 typedef struct RALFContext {
58 int32_t channel_data[2][4096];
60 int filter_params; ///< combined filter parameters for the current channel data
61 int filter_length; ///< length of the filter for the current channel data
62 int filter_bits; ///< filter precision for the current channel data
65 int bias[2]; ///< a constant value added to channel data after filtering
67 int num_blocks; ///< number of blocks inside the frame
69 int block_size[1 << 12]; ///< size of the blocks
70 int block_pts[1 << 12]; ///< block start time (in milliseconds)
76 #define MAX_ELEMS 644 // no RALF table uses more than that
78 static av_cold int init_ralf_vlc(VLC *vlc, const uint8_t *data, int elems)
80 uint8_t lens[MAX_ELEMS];
81 uint16_t codes[MAX_ELEMS];
82 int counts[17], prefixes[18];
87 for (i = 0; i <= 16; i++)
89 for (i = 0; i < elems; i++) {
90 cur_len = (nb ? *data & 0xF : *data >> 4) + 1;
92 max_bits = FFMAX(max_bits, cur_len);
98 for (i = 1; i <= 16; i++)
99 prefixes[i + 1] = (prefixes[i] + counts[i]) << 1;
101 for (i = 0; i < elems; i++)
102 codes[i] = prefixes[lens[i]]++;
104 return ff_init_vlc_sparse(vlc, FFMIN(max_bits, 9), elems,
105 lens, 1, 1, codes, 2, 2, NULL, 0, 0, 0);
108 static av_cold int decode_close(AVCodecContext *avctx)
110 RALFContext *ctx = avctx->priv_data;
113 for (i = 0; i < 3; i++) {
114 ff_free_vlc(&ctx->sets[i].filter_params);
115 ff_free_vlc(&ctx->sets[i].bias);
116 ff_free_vlc(&ctx->sets[i].coding_mode);
117 for (j = 0; j < 10; j++)
118 for (k = 0; k < 11; k++)
119 ff_free_vlc(&ctx->sets[i].filter_coeffs[j][k]);
120 for (j = 0; j < 15; j++)
121 ff_free_vlc(&ctx->sets[i].short_codes[j]);
122 for (j = 0; j < 125; j++)
123 ff_free_vlc(&ctx->sets[i].long_codes[j]);
129 static av_cold int decode_init(AVCodecContext *avctx)
131 RALFContext *ctx = avctx->priv_data;
135 if (avctx->extradata_size < 24 || memcmp(avctx->extradata, "LSD:", 4)) {
136 av_log(avctx, AV_LOG_ERROR, "Extradata is not groovy, dude\n");
137 return AVERROR_INVALIDDATA;
140 ctx->version = AV_RB16(avctx->extradata + 4);
141 if (ctx->version != 0x103) {
142 avpriv_request_sample(avctx, "Unknown version %X", ctx->version);
143 return AVERROR_PATCHWELCOME;
146 avctx->channels = AV_RB16(avctx->extradata + 8);
147 avctx->sample_rate = AV_RB32(avctx->extradata + 12);
148 if (avctx->channels < 1 || avctx->channels > 2
149 || avctx->sample_rate < 8000 || avctx->sample_rate > 96000) {
150 av_log(avctx, AV_LOG_ERROR, "Invalid coding parameters %d Hz %d ch\n",
151 avctx->sample_rate, avctx->channels);
152 return AVERROR_INVALIDDATA;
154 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
155 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO
158 ctx->max_frame_size = AV_RB32(avctx->extradata + 16);
159 if (ctx->max_frame_size > (1 << 20) || !ctx->max_frame_size) {
160 av_log(avctx, AV_LOG_ERROR, "invalid frame size %d\n",
161 ctx->max_frame_size);
163 ctx->max_frame_size = FFMAX(ctx->max_frame_size, avctx->sample_rate);
165 for (i = 0; i < 3; i++) {
166 ret = init_ralf_vlc(&ctx->sets[i].filter_params, filter_param_def[i],
167 FILTERPARAM_ELEMENTS);
172 ret = init_ralf_vlc(&ctx->sets[i].bias, bias_def[i], BIAS_ELEMENTS);
177 ret = init_ralf_vlc(&ctx->sets[i].coding_mode, coding_mode_def[i],
178 CODING_MODE_ELEMENTS);
183 for (j = 0; j < 10; j++) {
184 for (k = 0; k < 11; k++) {
185 ret = init_ralf_vlc(&ctx->sets[i].filter_coeffs[j][k],
186 filter_coeffs_def[i][j][k],
187 FILTER_COEFFS_ELEMENTS);
194 for (j = 0; j < 15; j++) {
195 ret = init_ralf_vlc(&ctx->sets[i].short_codes[j],
196 short_codes_def[i][j], SHORT_CODES_ELEMENTS);
202 for (j = 0; j < 125; j++) {
203 ret = init_ralf_vlc(&ctx->sets[i].long_codes[j],
204 long_codes_def[i][j], LONG_CODES_ELEMENTS);
215 static inline int extend_code(BitstreamContext *bc, int val, int range, int bits)
218 val = -range - get_ue_golomb(bc);
219 } else if (val == range * 2) {
220 val = range + get_ue_golomb(bc);
225 val = (val << bits) | bitstream_read(bc, bits);
229 static int decode_channel(RALFContext *ctx, BitstreamContext *bc, int ch,
230 int length, int mode, int bits)
234 VLCSet *set = ctx->sets + mode;
235 VLC *code_vlc; int range, range2, add_bits;
236 int *dst = ctx->channel_data[ch];
238 ctx->filter_params = bitstream_read_vlc(bc, set->filter_params.table, 9, 2);
239 ctx->filter_bits = (ctx->filter_params - 2) >> 6;
240 ctx->filter_length = ctx->filter_params - (ctx->filter_bits << 6) - 1;
242 if (ctx->filter_params == FILTER_RAW) {
243 for (i = 0; i < length; i++)
244 dst[i] = bitstream_read(bc, bits);
249 ctx->bias[ch] = bitstream_read_vlc(bc, set->bias.table, 9, 2);
250 ctx->bias[ch] = extend_code(bc, ctx->bias[ch], 127, 4);
252 if (ctx->filter_params == FILTER_NONE) {
253 memset(dst, 0, sizeof(*dst) * length);
257 if (ctx->filter_params > 1) {
258 int cmode = 0, coeff = 0;
259 VLC *vlc = set->filter_coeffs[ctx->filter_bits] + 5;
261 add_bits = ctx->filter_bits;
263 for (i = 0; i < ctx->filter_length; i++) {
264 t = bitstream_read_vlc(bc, vlc[cmode].table, vlc[cmode].bits, 2);
265 t = extend_code(bc, t, 21, add_bits);
267 coeff -= 12 << add_bits;
269 ctx->filter[i] = coeff;
271 cmode = coeff >> add_bits;
273 cmode = -1 - av_log2(-cmode);
276 } else if (cmode > 0) {
277 cmode = 1 + av_log2(cmode);
284 code_params = bitstream_read_vlc(bc, 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)
291 code_vlc = set->long_codes + code_params - 15;
296 code_vlc = set->short_codes + code_params;
299 for (i = 0; i < length; i += 2) {
302 t = bitstream_read_vlc(bc, code_vlc->table, code_vlc->bits, 2);
305 dst[i] = extend_code(bc, code1, range, 0) << add_bits;
306 dst[i + 1] = extend_code(bc, code2, range, 0) << add_bits;
308 dst[i] |= bitstream_read(bc, add_bits);
309 dst[i + 1] |= bitstream_read(bc, add_bits);
316 static void apply_lpc(RALFContext *ctx, int ch, int length, int bits)
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;
323 for (i = 1; i < length; i++) {
324 int flen = FFMIN(ctx->filter_length, i);
327 for (j = 0; j < flen; j++)
328 acc += ctx->filter[j] * audio[i - j - 1];
330 acc = (acc + bias - 1) >> ctx->filter_bits;
331 acc = FFMAX(acc, min_clip);
333 acc = (acc + bias) >> ctx->filter_bits;
334 acc = FFMIN(acc, max_clip);
340 static int decode_block(AVCodecContext *avctx, BitstreamContext *bc,
341 int16_t *dst0, int16_t *dst1)
343 RALFContext *ctx = avctx->priv_data;
345 int dmode, mode[2], bits[2];
349 len = 12 - get_unary(bc, 0, 6);
351 if (len <= 7) len ^= 1; // codes for length = 6 and 7 are swapped
354 if (ctx->sample_offset + len > ctx->max_frame_size) {
355 av_log(avctx, AV_LOG_ERROR,
356 "Decoder's stomach is crying, it ate too many samples\n");
357 return AVERROR_INVALIDDATA;
360 if (avctx->channels > 1)
361 dmode = bitstream_read(bc, 2) + 1;
365 mode[0] = (dmode == 4) ? 1 : 0;
366 mode[1] = (dmode >= 2) ? 2 : 0;
368 bits[1] = (mode[1] == 2) ? 17 : 16;
370 for (ch = 0; ch < avctx->channels; ch++) {
371 if ((ret = decode_channel(ctx, bc, ch, len, mode[ch], bits[ch])) < 0)
373 if (ctx->filter_params > 1 && ctx->filter_params != FILTER_RAW) {
374 ctx->filter_bits += 3;
375 apply_lpc(ctx, ch, len, bits[ch]);
377 if (bitstream_bits_left(bc) < 0)
378 return AVERROR_INVALIDDATA;
380 ch0 = ctx->channel_data[0];
381 ch1 = ctx->channel_data[1];
384 for (i = 0; i < len; i++)
385 dst0[i] = ch0[i] + ctx->bias[0];
388 for (i = 0; i < len; i++) {
389 dst0[i] = ch0[i] + ctx->bias[0];
390 dst1[i] = ch1[i] + ctx->bias[1];
394 for (i = 0; i < len; i++) {
395 ch0[i] += ctx->bias[0];
397 dst1[i] = ch0[i] - (ch1[i] + ctx->bias[1]);
401 for (i = 0; i < len; i++) {
402 t = ch0[i] + ctx->bias[0];
403 t2 = ch1[i] + ctx->bias[1];
409 for (i = 0; i < len; i++) {
410 t = ch1[i] + ctx->bias[1];
411 t2 = ((ch0[i] + ctx->bias[0]) << 1) | (t & 1);
412 dst0[i] = (t2 + t) / 2;
413 dst1[i] = (t2 - t) / 2;
418 ctx->sample_offset += len;
423 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
426 RALFContext *ctx = avctx->priv_data;
427 AVFrame *frame = data;
432 int table_size, table_bytes, i;
433 const uint8_t *src, *block_pointer;
439 table_bytes = (AV_RB16(avpkt->data) + 7) >> 3;
440 if (table_bytes + 3 > avpkt->size || avpkt->size > RALF_MAX_PKT_SIZE) {
441 av_log(avctx, AV_LOG_ERROR, "Wrong packet's breath smells of wrong data!\n");
442 return AVERROR_INVALIDDATA;
444 if (memcmp(ctx->pkt, avpkt->data, 2 + table_bytes)) {
445 av_log(avctx, AV_LOG_ERROR, "Wrong packet tails are wrong!\n");
446 return AVERROR_INVALIDDATA;
450 src_size = RALF_MAX_PKT_SIZE + avpkt->size;
451 memcpy(ctx->pkt + RALF_MAX_PKT_SIZE, avpkt->data + 2 + table_bytes,
452 avpkt->size - 2 - table_bytes);
454 if (avpkt->size == RALF_MAX_PKT_SIZE) {
455 memcpy(ctx->pkt, avpkt->data, avpkt->size);
462 src_size = avpkt->size;
465 frame->nb_samples = ctx->max_frame_size;
466 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
467 av_log(avctx, AV_LOG_ERROR, "Me fail get_buffer()? That's unpossible!\n");
470 samples0 = (int16_t *)frame->data[0];
471 samples1 = (int16_t *)frame->data[1];
474 av_log(avctx, AV_LOG_ERROR, "too short packets are too short!\n");
475 return AVERROR_INVALIDDATA;
477 table_size = AV_RB16(src);
478 table_bytes = (table_size + 7) >> 3;
479 if (src_size < table_bytes + 3) {
480 av_log(avctx, AV_LOG_ERROR, "short packets are short!\n");
481 return AVERROR_INVALIDDATA;
483 bitstream_init(&bc, src + 2, table_size);
485 while (bitstream_bits_left(&bc) > 0) {
486 ctx->block_size[ctx->num_blocks] = bitstream_read(&bc, 15);
487 if (bitstream_read_bit(&bc)) {
488 ctx->block_pts[ctx->num_blocks] = bitstream_read(&bc, 9);
490 ctx->block_pts[ctx->num_blocks] = 0;
495 block_pointer = src + table_bytes + 2;
496 bytes_left = src_size - table_bytes - 2;
497 ctx->sample_offset = 0;
498 for (i = 0; i < ctx->num_blocks; i++) {
499 if (bytes_left < ctx->block_size[i]) {
500 av_log(avctx, AV_LOG_ERROR, "I'm pedaling backwards\n");
503 bitstream_init8(&bc, block_pointer, ctx->block_size[i]);
504 if (decode_block(avctx, &bc, samples0 + ctx->sample_offset,
505 samples1 + ctx->sample_offset) < 0) {
506 av_log(avctx, AV_LOG_ERROR, "Sir, I got carsick in your office. Not decoding the rest of packet.\n");
509 block_pointer += ctx->block_size[i];
510 bytes_left -= ctx->block_size[i];
513 frame->nb_samples = ctx->sample_offset;
514 *got_frame_ptr = ctx->sample_offset > 0;
519 static void decode_flush(AVCodecContext *avctx)
521 RALFContext *ctx = avctx->priv_data;
527 AVCodec ff_ralf_decoder = {
529 .long_name = NULL_IF_CONFIG_SMALL("RealAudio Lossless"),
530 .type = AVMEDIA_TYPE_AUDIO,
531 .id = AV_CODEC_ID_RALF,
532 .priv_data_size = sizeof(RALFContext),
534 .close = decode_close,
535 .decode = decode_frame,
536 .flush = decode_flush,
537 .capabilities = AV_CODEC_CAP_DR1,
538 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
539 AV_SAMPLE_FMT_NONE },