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,
207 bytestream2_init(&gb, header, header_size);
209 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
210 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
211 return AVERROR_INVALIDDATA;
214 bytestream2_skip(&gb, 4); /* chunk size */
216 if (bytestream2_get_le32(&gb) != MKTAG('W', 'A', 'V', 'E')) {
217 av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
218 return AVERROR_INVALIDDATA;
221 while (bytestream2_get_le32(&gb) != MKTAG('f', 'm', 't', ' ')) {
222 len = bytestream2_get_le32(&gb);
223 bytestream2_skip(&gb, len);
224 if (bytestream2_get_bytes_left(&gb) < 16) {
225 av_log(avctx, AV_LOG_ERROR, "no fmt chunk found\n");
226 return AVERROR_INVALIDDATA;
229 len = bytestream2_get_le32(&gb);
232 av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
233 return AVERROR_INVALIDDATA;
236 wave_format = bytestream2_get_le16(&gb);
238 switch (wave_format) {
239 case WAVE_FORMAT_PCM:
242 av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
243 return AVERROR(ENOSYS);
246 bytestream2_skip(&gb, 2); // skip channels (already got from shorten header)
247 avctx->sample_rate = bytestream2_get_le32(&gb);
248 bytestream2_skip(&gb, 4); // skip bit rate (represents original uncompressed bit rate)
249 bytestream2_skip(&gb, 2); // skip block align (not needed)
250 avctx->bits_per_coded_sample = bytestream2_get_le16(&gb);
252 if (avctx->bits_per_coded_sample != 16) {
253 av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
254 return AVERROR(ENOSYS);
259 av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
264 static void output_buffer(int16_t **samples, int nchan, int blocksize,
268 for (ch = 0; ch < nchan; ch++) {
269 int32_t *in = buffer[ch];
270 int16_t *out = samples[ch];
271 for (i = 0; i < blocksize; i++)
272 out[i] = av_clip_int16(in[i]);
276 static const int fixed_coeffs[3][3] = {
282 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
283 int residual_size, int32_t coffset)
285 int pred_order, sum, qshift, init_sum, i, j;
288 if (command == FN_QLPC) {
289 /* read/validate prediction order */
290 pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
291 if (pred_order > s->nwrap) {
292 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
294 return AVERROR(EINVAL);
296 /* read LPC coefficients */
297 for (i = 0; i < pred_order; i++)
298 s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
303 /* fixed LPC coeffs */
304 pred_order = command;
305 coeffs = fixed_coeffs[pred_order - 1];
309 /* subtract offset from previous samples to use in prediction */
310 if (command == FN_QLPC && coffset)
311 for (i = -pred_order; i < 0; i++)
312 s->decoded[channel][i] -= coffset;
314 /* decode residual and do LPC prediction */
315 init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
316 for (i = 0; i < s->blocksize; i++) {
318 for (j = 0; j < pred_order; j++)
319 sum += coeffs[j] * s->decoded[channel][i - j - 1];
320 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
324 /* add offset to current samples */
325 if (command == FN_QLPC && coffset)
326 for (i = 0; i < s->blocksize; i++)
327 s->decoded[channel][i] += coffset;
332 static int read_header(ShortenContext *s)
336 /* shorten signature */
337 if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
338 av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
339 return AVERROR_INVALIDDATA;
343 s->blocksize = DEFAULT_BLOCK_SIZE;
345 s->version = get_bits(&s->gb, 8);
346 s->internal_ftype = get_uint(s, TYPESIZE);
348 s->channels = get_uint(s, CHANSIZE);
350 av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
351 return AVERROR_INVALIDDATA;
353 if (s->channels > MAX_CHANNELS) {
354 av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
356 return AVERROR_INVALIDDATA;
358 s->avctx->channels = s->channels;
360 /* get blocksize if version > 0 */
361 if (s->version > 0) {
365 blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
366 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
367 av_log(s->avctx, AV_LOG_ERROR,
368 "invalid or unsupported block size: %d\n",
370 return AVERROR(EINVAL);
372 s->blocksize = blocksize;
374 maxnlpc = get_uint(s, LPCQSIZE);
375 s->nmean = get_uint(s, 0);
377 skip_bytes = get_uint(s, NSKIPSIZE);
378 for (i = 0; i < skip_bytes; i++)
379 skip_bits(&s->gb, 8);
381 s->nwrap = FFMAX(NWRAP, maxnlpc);
383 if ((ret = allocate_buffers(s)) < 0)
386 if ((ret = init_offset(s)) < 0)
390 s->lpcqoffset = V2LPCQOFFSET;
392 if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
393 av_log(s->avctx, AV_LOG_ERROR,
394 "missing verbatim section at beginning of stream\n");
395 return AVERROR_INVALIDDATA;
398 s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
399 if (s->header_size >= OUT_BUFFER_SIZE ||
400 s->header_size < CANONICAL_HEADER_SIZE) {
401 av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
403 return AVERROR_INVALIDDATA;
406 for (i = 0; i < s->header_size; i++)
407 s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
409 if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
420 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
421 int *got_frame_ptr, AVPacket *avpkt)
423 AVFrame *frame = data;
424 const uint8_t *buf = avpkt->data;
425 int buf_size = avpkt->size;
426 ShortenContext *s = avctx->priv_data;
427 int i, input_buf_size = 0;
430 /* allocate internal bitstream buffer */
431 if (s->max_framesize == 0) {
433 s->max_framesize = 1024; // should hopefully be enough for the first header
434 tmp_ptr = av_fast_realloc(s->bitstream, &s->allocated_bitstream_size,
437 av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
438 return AVERROR(ENOMEM);
440 s->bitstream = tmp_ptr;
443 /* append current packet data to bitstream buffer */
444 if (1 && s->max_framesize) { //FIXME truncated
445 buf_size = FFMIN(buf_size, s->max_framesize - s->bitstream_size);
446 input_buf_size = buf_size;
448 if (s->bitstream_index + s->bitstream_size + buf_size >
449 s->allocated_bitstream_size) {
450 memmove(s->bitstream, &s->bitstream[s->bitstream_index],
452 s->bitstream_index = 0;
455 memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf,
457 buf = &s->bitstream[s->bitstream_index];
458 buf_size += s->bitstream_size;
459 s->bitstream_size = buf_size;
461 /* do not decode until buffer has at least max_framesize bytes or
462 * the end of the file has been reached */
463 if (buf_size < s->max_framesize && avpkt->data) {
465 return input_buf_size;
468 /* init and position bitstream reader */
469 init_get_bits(&s->gb, buf, buf_size * 8);
470 skip_bits(&s->gb, s->bitindex);
472 /* process header or next subblock */
473 if (!s->got_header) {
474 if ((ret = read_header(s)) < 0)
480 /* if quit command was read previously, don't decode anything */
481 if (s->got_quit_command) {
487 while (s->cur_chan < s->channels) {
491 if (get_bits_left(&s->gb) < 3 + FNSIZE) {
496 cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
498 if (cmd > FN_VERBATIM) {
499 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
504 if (!is_audio_command[cmd]) {
505 /* process non-audio command */
508 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
510 get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
513 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
516 unsigned blocksize = get_uint(s, av_log2(s->blocksize));
517 if (blocksize > s->blocksize) {
518 av_log(avctx, AV_LOG_ERROR,
519 "Increasing block size is not supported\n");
520 return AVERROR_PATCHWELCOME;
522 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
523 av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
524 "block size: %d\n", blocksize);
525 return AVERROR(EINVAL);
527 s->blocksize = blocksize;
531 s->got_quit_command = 1;
534 if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
539 /* process audio command */
540 int residual_size = 0;
541 int channel = s->cur_chan;
544 /* get Rice code for residual decoding */
545 if (cmd != FN_ZERO) {
546 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
547 /* This is a hack as version 0 differed in the definition
548 * of get_sr_golomb_shorten(). */
553 /* calculate sample offset using means from previous blocks */
555 coffset = s->offset[channel][0];
557 int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
558 for (i = 0; i < s->nmean; i++)
559 sum += s->offset[channel][i];
560 coffset = sum / s->nmean;
562 coffset >>= FFMIN(1, s->bitshift);
565 /* decode samples for this channel */
566 if (cmd == FN_ZERO) {
567 for (i = 0; i < s->blocksize; i++)
568 s->decoded[channel][i] = 0;
570 if ((ret = decode_subframe_lpc(s, cmd, channel,
571 residual_size, coffset)) < 0)
575 /* update means with info from the current block */
577 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
578 for (i = 0; i < s->blocksize; i++)
579 sum += s->decoded[channel][i];
581 for (i = 1; i < s->nmean; i++)
582 s->offset[channel][i - 1] = s->offset[channel][i];
585 s->offset[channel][s->nmean - 1] = sum / s->blocksize;
587 s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
590 /* copy wrap samples for use with next block */
591 for (i = -s->nwrap; i < 0; i++)
592 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
594 /* shift samples to add in unused zero bits which were removed
596 fix_bitshift(s, s->decoded[channel]);
598 /* if this is the last channel in the block, output the samples */
600 if (s->cur_chan == s->channels) {
601 /* get output buffer */
602 frame->nb_samples = s->blocksize;
603 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
604 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
607 /* interleave output */
608 output_buffer((int16_t **)frame->extended_data, s->channels,
609 s->blocksize, s->decoded);
615 if (s->cur_chan < s->channels)
619 s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
620 i = get_bits_count(&s->gb) / 8;
622 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
623 s->bitstream_size = 0;
624 s->bitstream_index = 0;
625 return AVERROR_INVALIDDATA;
627 if (s->bitstream_size) {
628 s->bitstream_index += i;
629 s->bitstream_size -= i;
630 return input_buf_size;
635 static av_cold int shorten_decode_close(AVCodecContext *avctx)
637 ShortenContext *s = avctx->priv_data;
640 for (i = 0; i < s->channels; i++) {
641 s->decoded[i] = NULL;
642 av_freep(&s->decoded_base[i]);
643 av_freep(&s->offset[i]);
645 av_freep(&s->bitstream);
646 av_freep(&s->coeffs);
651 AVCodec ff_shorten_decoder = {
653 .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
654 .type = AVMEDIA_TYPE_AUDIO,
655 .id = AV_CODEC_ID_SHORTEN,
656 .priv_data_size = sizeof(ShortenContext),
657 .init = shorten_decode_init,
658 .close = shorten_decode_close,
659 .decode = shorten_decode_frame,
660 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
661 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
662 AV_SAMPLE_FMT_NONE },