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");
131 if(s->blocksize + s->nwrap >= UINT_MAX/sizeof(int32_t) || s->blocksize + s->nwrap <= (unsigned)s->nwrap){
132 av_log(s->avctx, AV_LOG_ERROR, "s->blocksize + s->nwrap too large\n");
136 tmp_ptr = av_realloc(s->offset[chan], sizeof(int32_t)*FFMAX(1, s->nmean));
138 return AVERROR(ENOMEM);
139 s->offset[chan] = tmp_ptr;
141 tmp_ptr = av_realloc(s->decoded_base[chan], (s->blocksize + s->nwrap) *
142 sizeof(s->decoded_base[0][0]));
144 return AVERROR(ENOMEM);
145 s->decoded_base[chan] = tmp_ptr;
146 for (i=0; i<s->nwrap; i++)
147 s->decoded_base[chan][i] = 0;
148 s->decoded[chan] = s->decoded_base[chan] + s->nwrap;
151 coeffs = av_realloc(s->coeffs, s->nwrap * sizeof(*s->coeffs));
153 return AVERROR(ENOMEM);
160 static inline unsigned int get_uint(ShortenContext *s, int k)
163 k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
164 return get_ur_golomb_shorten(&s->gb, k);
168 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
172 if (s->bitshift != 0)
173 for (i = 0; i < s->blocksize; i++)
174 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)
191 av_log(s->avctx, AV_LOG_ERROR, "unknown audio type");
192 return AVERROR_INVALIDDATA;
195 for (chan = 0; chan < s->channels; chan++)
196 for (i = 0; i < nblock; i++)
197 s->offset[chan][i] = mean;
201 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
208 if (bytestream_get_le32(&header) != MKTAG('R','I','F','F')) {
209 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
213 header += 4; /* chunk size */;
215 if (bytestream_get_le32(&header) != MKTAG('W','A','V','E')) {
216 av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
220 while (bytestream_get_le32(&header) != MKTAG('f','m','t',' ')) {
221 len = bytestream_get_le32(&header);
224 len = bytestream_get_le32(&header);
227 av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
231 wave_format = bytestream_get_le16(&header);
233 switch (wave_format) {
234 case WAVE_FORMAT_PCM:
237 av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
241 header += 2; // skip channels (already got from shorten header)
242 avctx->sample_rate = bytestream_get_le32(&header);
243 header += 4; // skip bit rate (represents original uncompressed bit rate)
244 header += 2; // skip block align (not needed)
245 avctx->bits_per_coded_sample = bytestream_get_le16(&header);
247 if (avctx->bits_per_coded_sample != 16) {
248 av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample\n");
254 av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
259 static void output_buffer(int16_t **samples, int nchan, int blocksize,
263 for (ch = 0; ch < nchan; ch++) {
264 int32_t *in = buffer[ch];
265 int16_t *out = samples[ch];
266 for (i = 0; i < blocksize; i++)
267 out[i] = av_clip_int16(in[i]);
271 static const int fixed_coeffs[3][3] = {
277 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
278 int residual_size, int32_t coffset)
280 int pred_order, sum, qshift, init_sum, i, j;
283 if (command == FN_QLPC) {
284 /* read/validate prediction order */
285 pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
286 if (pred_order > s->nwrap) {
287 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n", pred_order);
288 return AVERROR(EINVAL);
290 /* read LPC coefficients */
291 for (i=0; i<pred_order; i++)
292 s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
297 /* fixed LPC coeffs */
298 pred_order = command;
299 coeffs = fixed_coeffs[pred_order-1];
303 /* subtract offset from previous samples to use in prediction */
304 if (command == FN_QLPC && coffset)
305 for (i = -pred_order; i < 0; i++)
306 s->decoded[channel][i] -= coffset;
308 /* decode residual and do LPC prediction */
309 init_sum = pred_order ? (command == FN_QLPC ? s->lpcqoffset : 0) : coffset;
310 for (i=0; i < s->blocksize; i++) {
312 for (j=0; j<pred_order; j++)
313 sum += coeffs[j] * s->decoded[channel][i-j-1];
314 s->decoded[channel][i] = get_sr_golomb_shorten(&s->gb, residual_size) + (sum >> qshift);
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");
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);
346 s->avctx->channels = s->channels;
348 /* get blocksize if version > 0 */
349 if (s->version > 0) {
350 int skip_bytes, blocksize;
352 blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
353 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
354 av_log(s->avctx, AV_LOG_ERROR, "invalid or unsupported block size: %d\n",
356 return AVERROR(EINVAL);
358 s->blocksize = blocksize;
360 maxnlpc = get_uint(s, LPCQSIZE);
361 s->nmean = get_uint(s, 0);
363 skip_bytes = get_uint(s, NSKIPSIZE);
364 for (i=0; i<skip_bytes; i++) {
365 skip_bits(&s->gb, 8);
368 s->nwrap = FFMAX(NWRAP, maxnlpc);
370 if ((ret = allocate_buffers(s)) < 0)
373 if ((ret = init_offset(s)) < 0)
377 s->lpcqoffset = V2LPCQOFFSET;
379 if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
380 av_log(s->avctx, AV_LOG_ERROR, "missing verbatim section at beginning of stream\n");
384 s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
385 if (s->header_size >= OUT_BUFFER_SIZE || s->header_size < CANONICAL_HEADER_SIZE) {
386 av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n", s->header_size);
390 for (i=0; i<s->header_size; i++)
391 s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
393 if (decode_wave_header(s->avctx, s->header, s->header_size) < 0)
404 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
405 int *got_frame_ptr, AVPacket *avpkt)
407 AVFrame *frame = data;
408 const uint8_t *buf = avpkt->data;
409 int buf_size = avpkt->size;
410 ShortenContext *s = avctx->priv_data;
411 int i, input_buf_size = 0;
414 /* allocate internal bitstream buffer */
415 if(s->max_framesize == 0){
417 s->max_framesize= 1024; // should hopefully be enough for the first header
418 tmp_ptr = av_fast_realloc(s->bitstream, &s->allocated_bitstream_size,
421 av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
422 return AVERROR(ENOMEM);
424 s->bitstream = tmp_ptr;
427 /* append current packet data to bitstream buffer */
428 if(1 && s->max_framesize){//FIXME truncated
429 buf_size= FFMIN(buf_size, s->max_framesize - s->bitstream_size);
430 input_buf_size= buf_size;
432 if(s->bitstream_index + s->bitstream_size + buf_size > s->allocated_bitstream_size){
433 memmove(s->bitstream, &s->bitstream[s->bitstream_index], s->bitstream_size);
434 s->bitstream_index=0;
437 memcpy(&s->bitstream[s->bitstream_index + s->bitstream_size], buf, buf_size);
438 buf= &s->bitstream[s->bitstream_index];
439 buf_size += s->bitstream_size;
440 s->bitstream_size= buf_size;
442 /* do not decode until buffer has at least max_framesize bytes or
443 the end of the file has been reached */
444 if (buf_size < s->max_framesize && avpkt->data) {
446 return input_buf_size;
449 /* init and position bitstream reader */
450 init_get_bits(&s->gb, buf, buf_size*8);
451 skip_bits(&s->gb, s->bitindex);
453 /* process header or next subblock */
454 if (!s->got_header) {
455 if ((ret = read_header(s)) < 0)
461 /* if quit command was read previously, don't decode anything */
462 if (s->got_quit_command) {
468 while (s->cur_chan < s->channels) {
472 if (get_bits_left(&s->gb) < 3+FNSIZE) {
477 cmd = get_ur_golomb_shorten(&s->gb, FNSIZE);
479 if (cmd > FN_VERBATIM) {
480 av_log(avctx, AV_LOG_ERROR, "unknown shorten function %d\n", cmd);
485 if (!is_audio_command[cmd]) {
486 /* process non-audio command */
489 len = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
491 get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
495 s->bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
498 int blocksize = get_uint(s, av_log2(s->blocksize));
499 if (blocksize > s->blocksize) {
500 av_log(avctx, AV_LOG_ERROR, "Increasing block size is not supported\n");
501 return AVERROR_PATCHWELCOME;
503 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
504 av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
505 "block size: %d\n", blocksize);
506 return AVERROR(EINVAL);
508 s->blocksize = blocksize;
512 s->got_quit_command = 1;
515 if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
520 /* process audio command */
521 int residual_size = 0;
522 int channel = s->cur_chan;
525 /* get Rice code for residual decoding */
526 if (cmd != FN_ZERO) {
527 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
528 /* This is a hack as version 0 differed in the definition
529 * of get_sr_golomb_shorten(). */
534 /* calculate sample offset using means from previous blocks */
536 coffset = s->offset[channel][0];
538 int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
539 for (i=0; i<s->nmean; i++)
540 sum += s->offset[channel][i];
541 coffset = sum / s->nmean;
543 coffset >>= FFMIN(1, s->bitshift);
546 /* decode samples for this channel */
547 if (cmd == FN_ZERO) {
548 for (i=0; i<s->blocksize; i++)
549 s->decoded[channel][i] = 0;
551 if ((ret = decode_subframe_lpc(s, cmd, channel, residual_size, coffset)) < 0)
555 /* update means with info from the current block */
557 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
558 for (i=0; i<s->blocksize; i++)
559 sum += s->decoded[channel][i];
561 for (i=1; i<s->nmean; i++)
562 s->offset[channel][i-1] = s->offset[channel][i];
565 s->offset[channel][s->nmean - 1] = sum / s->blocksize;
567 s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
570 /* copy wrap samples for use with next block */
571 for (i=-s->nwrap; i<0; i++)
572 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
574 /* shift samples to add in unused zero bits which were removed
576 fix_bitshift(s, s->decoded[channel]);
578 /* if this is the last channel in the block, output the samples */
580 if (s->cur_chan == s->channels) {
581 /* get output buffer */
582 frame->nb_samples = s->blocksize;
583 if ((ret = ff_get_buffer(avctx, frame)) < 0) {
584 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
587 /* interleave output */
588 output_buffer((int16_t **)frame->extended_data, s->channels,
589 s->blocksize, s->decoded);
595 if (s->cur_chan < s->channels)
599 s->bitindex = get_bits_count(&s->gb) - 8*((get_bits_count(&s->gb))/8);
600 i= (get_bits_count(&s->gb))/8;
602 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
604 s->bitstream_index=0;
607 if (s->bitstream_size) {
608 s->bitstream_index += i;
609 s->bitstream_size -= i;
610 return input_buf_size;
615 static av_cold int shorten_decode_close(AVCodecContext *avctx)
617 ShortenContext *s = avctx->priv_data;
620 for (i = 0; i < s->channels; i++) {
621 s->decoded[i] = NULL;
622 av_freep(&s->decoded_base[i]);
623 av_freep(&s->offset[i]);
625 av_freep(&s->bitstream);
626 av_freep(&s->coeffs);
631 AVCodec ff_shorten_decoder = {
633 .type = AVMEDIA_TYPE_AUDIO,
634 .id = AV_CODEC_ID_SHORTEN,
635 .priv_data_size = sizeof(ShortenContext),
636 .init = shorten_decode_init,
637 .close = shorten_decode_close,
638 .decode = shorten_decode_frame,
639 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
640 .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
641 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
642 AV_SAMPLE_FMT_NONE },