3 * Copyright (c) 2005 Jeff Muizelaar
5 * This file is part of FFmpeg.
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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
64 #define V2LPCQOFFSET (1 << LPCQUANT)
72 #define FN_BLOCKSIZE 5
78 /** indicates if the FN_* command is audio or non-audio */
79 static const uint8_t is_audio_command[10] = { 1, 1, 1, 1, 0, 0, 0, 1, 1, 0 };
81 #define VERBATIM_CKSIZE_SIZE 5
82 #define VERBATIM_BYTE_SIZE 8
83 #define CANONICAL_HEADER_SIZE 44
85 typedef struct ShortenContext {
86 AVCodecContext *avctx;
89 int min_framesize, max_framesize;
92 int32_t *decoded[MAX_CHANNELS];
93 int32_t *decoded_base[MAX_CHANNELS];
94 int32_t *offset[MAX_CHANNELS];
99 unsigned int allocated_bitstream_size;
101 uint8_t header[OUT_BUFFER_SIZE];
112 int got_quit_command;
115 static av_cold int shorten_decode_init(AVCodecContext *avctx)
117 ShortenContext *s = avctx->priv_data;
123 static int allocate_buffers(ShortenContext *s)
129 for (chan = 0; chan < s->channels; chan++) {
130 if (FFMAX(1, s->nmean) >= UINT_MAX / sizeof(int32_t)) {
131 av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
132 return AVERROR_INVALIDDATA;
134 if (s->blocksize + s->nwrap >= UINT_MAX / sizeof(int32_t) ||
135 s->blocksize + s->nwrap <= (unsigned)s->nwrap) {
136 av_log(s->avctx, AV_LOG_ERROR,
137 "s->blocksize + s->nwrap too large\n");
138 return AVERROR_INVALIDDATA;
142 av_realloc(s->offset[chan], sizeof(int32_t) * FFMAX(1, s->nmean));
144 return AVERROR(ENOMEM);
145 s->offset[chan] = tmp_ptr;
147 tmp_ptr = av_realloc(s->decoded_base[chan], (s->blocksize + s->nwrap) *
148 sizeof(s->decoded_base[0][0]));
150 return AVERROR(ENOMEM);
151 s->decoded_base[chan] = tmp_ptr;
152 for (i = 0; i < s->nwrap; i++)
153 s->decoded_base[chan][i] = 0;
154 s->decoded[chan] = s->decoded_base[chan] + s->nwrap;
157 coeffs = av_realloc(s->coeffs, s->nwrap * sizeof(*s->coeffs));
159 return AVERROR(ENOMEM);
165 static inline unsigned int get_uint(ShortenContext *s, int k)
168 k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
169 return get_ur_golomb_shorten(&s->gb, k);
172 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
176 if (s->bitshift != 0)
177 for (i = 0; i < s->blocksize; i++)
178 buffer[i] <<= s->bitshift;
181 static int init_offset(ShortenContext *s)
185 int nblock = FFMAX(1, s->nmean);
186 /* initialise offset */
187 switch (s->internal_ftype) {
189 s->avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
194 s->avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
197 av_log(s->avctx, AV_LOG_ERROR, "unknown audio type\n");
198 return AVERROR_PATCHWELCOME;
201 for (chan = 0; chan < s->channels; chan++)
202 for (i = 0; i < nblock; i++)
203 s->offset[chan][i] = mean;
207 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
212 const uint8_t *end= header + header_size;
214 if (bytestream_get_le32(&header) != MKTAG('R', 'I', 'F', 'F')) {
215 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
216 return AVERROR_INVALIDDATA;
219 header += 4; /* chunk size */
221 if (bytestream_get_le32(&header) != MKTAG('W', 'A', 'V', 'E')) {
222 av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
223 return AVERROR_INVALIDDATA;
226 while (bytestream_get_le32(&header) != MKTAG('f', 'm', 't', ' ')) {
227 len = bytestream_get_le32(&header);
228 if (len<0 || end - header - 8 < len)
229 return AVERROR_INVALIDDATA;
232 len = bytestream_get_le32(&header);
235 av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
236 return AVERROR_INVALIDDATA;
239 wave_format = bytestream_get_le16(&header);
241 switch (wave_format) {
242 case WAVE_FORMAT_PCM:
245 av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
246 return AVERROR(ENOSYS);
249 header += 2; // skip channels (already got from shorten header)
250 avctx->sample_rate = bytestream_get_le32(&header);
251 header += 4; // skip bit rate (represents original uncompressed bit rate)
252 header += 2; // skip block align (not needed)
253 bps = bytestream_get_le16(&header);
254 avctx->bits_per_coded_sample = bps;
256 if (bps != 16 && bps != 8) {
257 av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample: %d\n", bps);
258 return AVERROR(ENOSYS);
263 av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
268 static const int fixed_coeffs[3][3] = {
274 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
275 int residual_size, int32_t coffset)
277 int pred_order, sum, qshift, init_sum, i, j;
280 if (command == FN_QLPC) {
281 /* read/validate prediction order */
282 pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
283 if (pred_order > s->nwrap) {
284 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
286 return AVERROR(EINVAL);
288 /* read LPC coefficients */
289 for (i = 0; i < pred_order; i++)
290 s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
295 /* fixed LPC coeffs */
296 pred_order = command;
297 coeffs = fixed_coeffs[pred_order - 1];
301 /* subtract offset from previous samples to use in prediction */
302 if (command == FN_QLPC && coffset)
303 for (i = -pred_order; i < 0; i++)
304 s->decoded[channel][i] -= coffset;
306 /* decode residual and do LPC prediction */
307 init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
308 for (i = 0; i < s->blocksize; i++) {
310 for (j = 0; j < pred_order; j++)
311 sum += coeffs[j] * s->decoded[channel][i - j - 1];
312 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) +
316 /* add offset to current samples */
317 if (command == FN_QLPC && coffset)
318 for (i = 0; i < s->blocksize; i++)
319 s->decoded[channel][i] += coffset;
324 static int read_header(ShortenContext *s)
328 /* shorten signature */
329 if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
330 av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
331 return AVERROR_INVALIDDATA;
335 s->blocksize = DEFAULT_BLOCK_SIZE;
337 s->version = get_bits(&s->gb, 8);
338 s->internal_ftype = get_uint(s, TYPESIZE);
340 s->channels = get_uint(s, CHANSIZE);
342 av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
343 return AVERROR_INVALIDDATA;
345 if (s->channels > MAX_CHANNELS) {
346 av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
348 return AVERROR_INVALIDDATA;
350 s->avctx->channels = s->channels;
352 /* get blocksize if version > 0 */
353 if (s->version > 0) {
357 blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
358 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
359 av_log(s->avctx, AV_LOG_ERROR,
360 "invalid or unsupported block size: %d\n",
362 return AVERROR(EINVAL);
364 s->blocksize = blocksize;
366 maxnlpc = get_uint(s, LPCQSIZE);
367 s->nmean = get_uint(s, 0);
369 skip_bytes = get_uint(s, NSKIPSIZE);
370 for (i = 0; i < skip_bytes; i++)
371 skip_bits(&s->gb, 8);
373 s->nwrap = FFMAX(NWRAP, maxnlpc);
375 if ((ret = allocate_buffers(s)) < 0)
378 if ((ret = init_offset(s)) < 0)
382 s->lpcqoffset = V2LPCQOFFSET;
384 if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
385 av_log(s->avctx, AV_LOG_ERROR,
386 "missing verbatim section at beginning of stream\n");
387 return AVERROR_INVALIDDATA;
390 s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
391 if (s->header_size >= OUT_BUFFER_SIZE ||
392 s->header_size < CANONICAL_HEADER_SIZE) {
393 av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
395 return AVERROR_INVALIDDATA;
398 for (i = 0; i < s->header_size; i++)
399 s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
401 if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
412 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
413 int *got_frame_ptr, AVPacket *avpkt)
415 AVFrame *frame = data;
416 const uint8_t *buf = avpkt->data;
417 int buf_size = avpkt->size;
418 ShortenContext *s = avctx->priv_data;
419 int i, input_buf_size = 0;
422 /* allocate internal bitstream buffer */
423 if (s->max_framesize == 0) {
425 s->max_framesize = 8192; // should hopefully be enough for the first header
426 tmp_ptr = av_fast_realloc(s->bitstream, &s->allocated_bitstream_size,
429 av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
430 return AVERROR(ENOMEM);
432 s->bitstream = tmp_ptr;
435 /* append current packet data to bitstream buffer */
436 if (1 && s->max_framesize) { //FIXME truncated
437 buf_size = FFMIN(buf_size, s->max_framesize - s->bitstream_size);
438 input_buf_size = buf_size;
440 if (s->bitstream_index + s->bitstream_size + buf_size >
441 s->allocated_bitstream_size) {
442 memmove(s->bitstream, &s->bitstream[s->bitstream_index],
444 s->bitstream_index = 0;
447 memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf,
449 buf = &s->bitstream[s->bitstream_index];
450 buf_size += s->bitstream_size;
451 s->bitstream_size = buf_size;
453 /* do not decode until buffer has at least max_framesize bytes or
454 * the end of the file has been reached */
455 if (buf_size < s->max_framesize && avpkt->data) {
457 return input_buf_size;
460 /* init and position bitstream reader */
461 init_get_bits(&s->gb, buf, buf_size * 8);
462 skip_bits(&s->gb, s->bitindex);
464 /* process header or next subblock */
465 if (!s->got_header) {
466 if ((ret = read_header(s)) < 0)
472 /* if quit command was read previously, don't decode anything */
473 if (s->got_quit_command) {
479 while (s->cur_chan < s->channels) {
483 if (get_bits_left(&s->gb) < 3 + FNSIZE) {
488 cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
490 if (cmd > FN_VERBATIM) {
491 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
496 if (!is_audio_command[cmd]) {
497 /* process non-audio command */
500 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
502 get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
505 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
508 unsigned blocksize = get_uint(s, av_log2(s->blocksize));
509 if (blocksize > s->blocksize) {
510 av_log(avctx, AV_LOG_ERROR,
511 "Increasing block size is not supported\n");
512 return AVERROR_PATCHWELCOME;
514 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
515 av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
516 "block size: %d\n", blocksize);
517 return AVERROR(EINVAL);
519 s->blocksize = blocksize;
523 s->got_quit_command = 1;
526 if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
531 /* process audio command */
532 int residual_size = 0;
533 int channel = s->cur_chan;
536 /* get Rice code for residual decoding */
537 if (cmd != FN_ZERO) {
538 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
539 /* This is a hack as version 0 differed in the definition
540 * of get_sr_golomb_shorten(). */
545 /* calculate sample offset using means from previous blocks */
547 coffset = s->offset[channel][0];
549 int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
550 for (i = 0; i < s->nmean; i++)
551 sum += s->offset[channel][i];
552 coffset = sum / s->nmean;
554 coffset = s->bitshift == 0 ? coffset : coffset >> s->bitshift - 1 >> 1;
557 /* decode samples for this channel */
558 if (cmd == FN_ZERO) {
559 for (i = 0; i < s->blocksize; i++)
560 s->decoded[channel][i] = 0;
562 if ((ret = decode_subframe_lpc(s, cmd, channel,
563 residual_size, coffset)) < 0)
567 /* update means with info from the current block */
569 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
570 for (i = 0; i < s->blocksize; i++)
571 sum += s->decoded[channel][i];
573 for (i = 1; i < s->nmean; i++)
574 s->offset[channel][i - 1] = s->offset[channel][i];
577 s->offset[channel][s->nmean - 1] = sum / s->blocksize;
579 s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
582 /* copy wrap samples for use with next block */
583 for (i = -s->nwrap; i < 0; i++)
584 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
586 /* shift samples to add in unused zero bits which were removed
588 fix_bitshift(s, s->decoded[channel]);
590 /* if this is the last channel in the block, output the samples */
592 if (s->cur_chan == s->channels) {
594 int16_t *samples_s16;
597 /* get output buffer */
598 frame->nb_samples = s->blocksize;
599 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
602 for (chan = 0; chan < s->channels; chan++) {
603 samples_u8 = ((uint8_t **)frame->extended_data)[chan];
604 samples_s16 = ((int16_t **)frame->extended_data)[chan];
605 for (i = 0; i < s->blocksize; i++) {
606 switch (s->internal_ftype) {
608 *samples_u8++ = av_clip_uint8(s->decoded[chan][i]);
612 *samples_s16++ = av_clip_int16(s->decoded[chan][i]);
622 if (s->cur_chan < s->channels)
626 s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
627 i = get_bits_count(&s->gb) / 8;
629 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
630 s->bitstream_size = 0;
631 s->bitstream_index = 0;
632 return AVERROR_INVALIDDATA;
634 if (s->bitstream_size) {
635 s->bitstream_index += i;
636 s->bitstream_size -= i;
637 return input_buf_size;
642 static av_cold int shorten_decode_close(AVCodecContext *avctx)
644 ShortenContext *s = avctx->priv_data;
647 for (i = 0; i < s->channels; i++) {
648 s->decoded[i] = NULL;
649 av_freep(&s->decoded_base[i]);
650 av_freep(&s->offset[i]);
652 av_freep(&s->bitstream);
653 av_freep(&s->coeffs);
658 AVCodec ff_shorten_decoder = {
660 .type = AVMEDIA_TYPE_AUDIO,
661 .id = AV_CODEC_ID_SHORTEN,
662 .priv_data_size = sizeof(ShortenContext),
663 .init = shorten_decode_init,
664 .close = shorten_decode_close,
665 .decode = shorten_decode_frame,
666 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
667 .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
668 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
670 AV_SAMPLE_FMT_NONE },