3 * Copyright (c) 2005 Jeff Muizelaar
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * @author Jeff Muizelaar
31 #include "bytestream.h"
36 #define MAX_CHANNELS 8
37 #define MAX_BLOCKSIZE 65535
39 #define OUT_BUFFER_SIZE 16384
43 #define WAVE_FORMAT_PCM 0x0001
45 #define DEFAULT_BLOCK_SIZE 256
51 #define BITSHIFTSIZE 2
60 #define V2LPCQOFFSET (1 << LPCQUANT)
68 #define FN_BLOCKSIZE 5
74 /** indicates if the FN_* command is audio or non-audio */
75 static const uint8_t is_audio_command[10] = { 1, 1, 1, 1, 0, 0, 0, 1, 1, 0 };
77 #define VERBATIM_CKSIZE_SIZE 5
78 #define VERBATIM_BYTE_SIZE 8
79 #define CANONICAL_HEADER_SIZE 44
81 typedef struct ShortenContext {
82 AVCodecContext *avctx;
85 int min_framesize, max_framesize;
88 int32_t *decoded[MAX_CHANNELS];
89 int32_t *decoded_base[MAX_CHANNELS];
90 int32_t *offset[MAX_CHANNELS];
95 unsigned int allocated_bitstream_size;
97 uint8_t header[OUT_BUFFER_SIZE];
108 int got_quit_command;
111 static av_cold int shorten_decode_init(AVCodecContext *avctx)
113 ShortenContext *s = avctx->priv_data;
115 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
120 static int allocate_buffers(ShortenContext *s)
126 for (chan = 0; chan < s->channels; chan++) {
127 if (FFMAX(1, s->nmean) >= UINT_MAX / sizeof(int32_t)) {
128 av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
129 return AVERROR_INVALIDDATA;
131 if (s->blocksize + s->nwrap >= UINT_MAX / sizeof(int32_t) ||
132 s->blocksize + s->nwrap <= (unsigned)s->nwrap) {
133 av_log(s->avctx, AV_LOG_ERROR,
134 "s->blocksize + s->nwrap too large\n");
135 return AVERROR_INVALIDDATA;
139 av_realloc(s->offset[chan], sizeof(int32_t) * FFMAX(1, s->nmean));
141 return AVERROR(ENOMEM);
142 s->offset[chan] = tmp_ptr;
144 tmp_ptr = av_realloc(s->decoded_base[chan], (s->blocksize + s->nwrap) *
145 sizeof(s->decoded_base[0][0]));
147 return AVERROR(ENOMEM);
148 s->decoded_base[chan] = tmp_ptr;
149 for (i = 0; i < s->nwrap; i++)
150 s->decoded_base[chan][i] = 0;
151 s->decoded[chan] = s->decoded_base[chan] + s->nwrap;
154 coeffs = av_realloc(s->coeffs, s->nwrap * sizeof(*s->coeffs));
156 return AVERROR(ENOMEM);
162 static inline unsigned int get_uint(ShortenContext *s, int k)
165 k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
166 return get_ur_golomb_shorten(&s->gb, k);
169 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
173 if (s->bitshift != 0)
174 for (i = 0; i < s->blocksize; i++)
175 buffer[i] <<= s->bitshift;
178 static int init_offset(ShortenContext *s)
182 int nblock = FFMAX(1, s->nmean);
183 /* initialise offset */
184 switch (s->internal_ftype) {
190 av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
191 return AVERROR_INVALIDDATA;
194 for (chan = 0; chan < s->channels; chan++)
195 for (i = 0; i < nblock; i++)
196 s->offset[chan][i] = mean;
200 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
206 if (bytestream_get_le32(&header) != MKTAG('R', 'I', 'F', 'F')) {
207 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
208 return AVERROR_INVALIDDATA;
211 header += 4; /* chunk size */
213 if (bytestream_get_le32(&header) != MKTAG('W', 'A', 'V', 'E')) {
214 av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
215 return AVERROR_INVALIDDATA;
218 while (bytestream_get_le32(&header) != MKTAG('f', 'm', 't', ' ')) {
219 len = bytestream_get_le32(&header);
222 len = bytestream_get_le32(&header);
225 av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
226 return AVERROR_INVALIDDATA;
229 wave_format = bytestream_get_le16(&header);
231 switch (wave_format) {
232 case WAVE_FORMAT_PCM:
235 av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
236 return AVERROR(ENOSYS);
239 header += 2; // skip channels (already got from shorten header)
240 avctx->sample_rate = bytestream_get_le32(&header);
241 header += 4; // skip bit rate (represents original uncompressed bit rate)
242 header += 2; // skip block align (not needed)
243 avctx->bits_per_coded_sample = bytestream_get_le16(&header);
245 if (avctx->bits_per_coded_sample != 16) {
246 av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
247 return AVERROR(ENOSYS);
252 av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
257 static void output_buffer(int16_t **samples, int nchan, int blocksize,
261 for (ch = 0; ch < nchan; ch++) {
262 int32_t *in = buffer[ch];
263 int16_t *out = samples[ch];
264 for (i = 0; i < blocksize; i++)
265 out[i] = av_clip_int16(in[i]);
269 static const int fixed_coeffs[3][3] = {
275 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
276 int residual_size, int32_t coffset)
278 int pred_order, sum, qshift, init_sum, i, j;
281 if (command == FN_QLPC) {
282 /* read/validate prediction order */
283 pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
284 if (pred_order > s->nwrap) {
285 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
287 return AVERROR(EINVAL);
289 /* read LPC coefficients */
290 for (i = 0; i < pred_order; i++)
291 s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
296 /* fixed LPC coeffs */
297 pred_order = command;
298 coeffs = fixed_coeffs[pred_order - 1];
302 /* subtract offset from previous samples to use in prediction */
303 if (command == FN_QLPC && coffset)
304 for (i = -pred_order; i < 0; i++)
305 s->decoded[channel][i] -= coffset;
307 /* decode residual and do LPC prediction */
308 init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
309 for (i = 0; i < s->blocksize; i++) {
311 for (j = 0; j < pred_order; j++)
312 sum += coeffs[j] * s->decoded[channel][i - j - 1];
313 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
317 /* add offset to current samples */
318 if (command == FN_QLPC && coffset)
319 for (i = 0; i < s->blocksize; i++)
320 s->decoded[channel][i] += coffset;
325 static int read_header(ShortenContext *s)
329 /* shorten signature */
330 if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
331 av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
332 return AVERROR_INVALIDDATA;
336 s->blocksize = DEFAULT_BLOCK_SIZE;
338 s->version = get_bits(&s->gb, 8);
339 s->internal_ftype = get_uint(s, TYPESIZE);
341 s->channels = get_uint(s, CHANSIZE);
342 if (s->channels <= 0 || s->channels > MAX_CHANNELS) {
343 av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
345 return AVERROR_INVALIDDATA;
347 s->avctx->channels = s->channels;
349 /* get blocksize if version > 0 */
350 if (s->version > 0) {
351 int skip_bytes, blocksize;
353 blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
354 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
355 av_log(s->avctx, AV_LOG_ERROR,
356 "invalid or unsupported block size: %d\n",
358 return AVERROR(EINVAL);
360 s->blocksize = blocksize;
362 maxnlpc = get_uint(s, LPCQSIZE);
363 s->nmean = get_uint(s, 0);
365 skip_bytes = get_uint(s, NSKIPSIZE);
366 for (i = 0; i < skip_bytes; i++)
367 skip_bits(&s->gb, 8);
369 s->nwrap = FFMAX(NWRAP, maxnlpc);
371 if ((ret = allocate_buffers(s)) < 0)
374 if ((ret = init_offset(s)) < 0)
378 s->lpcqoffset = V2LPCQOFFSET;
380 if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
381 av_log(s->avctx, AV_LOG_ERROR,
382 "missing verbatim section at beginning of stream\n");
383 return AVERROR_INVALIDDATA;
386 s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
387 if (s->header_size >= OUT_BUFFER_SIZE ||
388 s->header_size < CANONICAL_HEADER_SIZE) {
389 av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
391 return AVERROR_INVALIDDATA;
394 for (i = 0; i < s->header_size; i++)
395 s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
397 if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
408 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
409 int *got_frame_ptr, AVPacket *avpkt)
411 AVFrame *frame = data;
412 const uint8_t *buf = avpkt->data;
413 int buf_size = avpkt->size;
414 ShortenContext *s = avctx->priv_data;
415 int i, input_buf_size = 0;
418 /* allocate internal bitstream buffer */
419 if (s->max_framesize == 0) {
421 s->max_framesize = 1024; // should hopefully be enough for the first header
422 tmp_ptr = av_fast_realloc(s->bitstream, &s->allocated_bitstream_size,
425 av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
426 return AVERROR(ENOMEM);
428 s->bitstream = tmp_ptr;
431 /* append current packet data to bitstream buffer */
432 if (1 && s->max_framesize) { //FIXME truncated
433 buf_size = FFMIN(buf_size, s->max_framesize - s->bitstream_size);
434 input_buf_size = buf_size;
436 if (s->bitstream_index + s->bitstream_size + buf_size >
437 s->allocated_bitstream_size) {
438 memmove(s->bitstream, &s->bitstream[s->bitstream_index],
440 s->bitstream_index = 0;
443 memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf,
445 buf = &s->bitstream[s->bitstream_index];
446 buf_size += s->bitstream_size;
447 s->bitstream_size = buf_size;
449 /* do not decode until buffer has at least max_framesize bytes or
450 * the end of the file has been reached */
451 if (buf_size < s->max_framesize && avpkt->data) {
453 return input_buf_size;
456 /* init and position bitstream reader */
457 init_get_bits(&s->gb, buf, buf_size * 8);
458 skip_bits(&s->gb, s->bitindex);
460 /* process header or next subblock */
461 if (!s->got_header) {
462 if ((ret = read_header(s)) < 0)
468 /* if quit command was read previously, don't decode anything */
469 if (s->got_quit_command) {
475 while (s->cur_chan < s->channels) {
479 if (get_bits_left(&s->gb) < 3 + FNSIZE) {
484 cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
486 if (cmd > FN_VERBATIM) {
487 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
492 if (!is_audio_command[cmd]) {
493 /* process non-audio command */
496 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
498 get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
501 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
504 int blocksize = get_uint(s, av_log2(s->blocksize));
505 if (blocksize > s->blocksize) {
506 av_log(avctx, AV_LOG_ERROR,
507 "Increasing block size is not supported\n");
508 return AVERROR_PATCHWELCOME;
510 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
511 av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
512 "block size: %d\n", blocksize);
513 return AVERROR(EINVAL);
515 s->blocksize = blocksize;
519 s->got_quit_command = 1;
522 if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
527 /* process audio command */
528 int residual_size = 0;
529 int channel = s->cur_chan;
532 /* get Rice code for residual decoding */
533 if (cmd != FN_ZERO) {
534 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
535 /* This is a hack as version 0 differed in the definition
536 * of get_sr_golomb_shorten(). */
541 /* calculate sample offset using means from previous blocks */
543 coffset = s->offset[channel][0];
545 int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
546 for (i = 0; i < s->nmean; i++)
547 sum += s->offset[channel][i];
548 coffset = sum / s->nmean;
550 coffset >>= FFMIN(1, s->bitshift);
553 /* decode samples for this channel */
554 if (cmd == FN_ZERO) {
555 for (i = 0; i < s->blocksize; i++)
556 s->decoded[channel][i] = 0;
558 if ((ret = decode_subframe_lpc(s, cmd, channel,
559 residual_size, coffset)) < 0)
563 /* update means with info from the current block */
565 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
566 for (i = 0; i < s->blocksize; i++)
567 sum += s->decoded[channel][i];
569 for (i = 1; i < s->nmean; i++)
570 s->offset[channel][i - 1] = s->offset[channel][i];
573 s->offset[channel][s->nmean - 1] = sum / s->blocksize;
575 s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
578 /* copy wrap samples for use with next block */
579 for (i = -s->nwrap; i < 0; i++)
580 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
582 /* shift samples to add in unused zero bits which were removed
584 fix_bitshift(s, s->decoded[channel]);
586 /* if this is the last channel in the block, output the samples */
588 if (s->cur_chan == s->channels) {
589 /* get output buffer */
590 frame->nb_samples = s->blocksize;
591 if ((ret = ff_get_buffer(avctx, frame)) < 0) {
592 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
595 /* interleave output */
596 output_buffer((int16_t **)frame->extended_data, s->channels,
597 s->blocksize, s->decoded);
603 if (s->cur_chan < s->channels)
607 s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
608 i = get_bits_count(&s->gb) / 8;
610 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
611 s->bitstream_size = 0;
612 s->bitstream_index = 0;
613 return AVERROR_INVALIDDATA;
615 if (s->bitstream_size) {
616 s->bitstream_index += i;
617 s->bitstream_size -= i;
618 return input_buf_size;
623 static av_cold int shorten_decode_close(AVCodecContext *avctx)
625 ShortenContext *s = avctx->priv_data;
628 for (i = 0; i < s->channels; i++) {
629 s->decoded[i] = NULL;
630 av_freep(&s->decoded_base[i]);
631 av_freep(&s->offset[i]);
633 av_freep(&s->bitstream);
634 av_freep(&s->coeffs);
639 AVCodec ff_shorten_decoder = {
641 .type = AVMEDIA_TYPE_AUDIO,
642 .id = AV_CODEC_ID_SHORTEN,
643 .priv_data_size = sizeof(ShortenContext),
644 .init = shorten_decode_init,
645 .close = shorten_decode_close,
646 .decode = shorten_decode_frame,
647 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
648 .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
649 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
650 AV_SAMPLE_FMT_NONE },