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)
127 for (chan = 0; chan < s->channels; chan++) {
128 if (FFMAX(1, s->nmean) >= UINT_MAX / sizeof(int32_t)) {
129 av_log(s->avctx, AV_LOG_ERROR, "nmean too large\n");
130 return AVERROR_INVALIDDATA;
132 if (s->blocksize + (uint64_t)s->nwrap >= UINT_MAX / sizeof(int32_t)) {
133 av_log(s->avctx, AV_LOG_ERROR,
134 "s->blocksize + s->nwrap too large\n");
135 return AVERROR_INVALIDDATA;
138 if ((err = av_reallocp_array(&s->offset[chan],
140 FFMAX(1, s->nmean))) < 0)
143 if ((err = av_reallocp_array(&s->decoded_base[chan], (s->blocksize + s->nwrap),
144 sizeof(s->decoded_base[0][0]))) < 0)
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 if ((err = av_reallocp_array(&s->coeffs, s->nwrap, sizeof(*s->coeffs))) < 0)
157 static inline unsigned int get_uint(ShortenContext *s, int k)
160 k = get_ur_golomb_shorten(&s->gb, ULONGSIZE);
161 return get_ur_golomb_shorten(&s->gb, k);
164 static void fix_bitshift(ShortenContext *s, int32_t *buffer)
168 if (s->bitshift != 0)
169 for (i = 0; i < s->blocksize; i++)
170 buffer[i] <<= s->bitshift;
173 static int init_offset(ShortenContext *s)
177 int nblock = FFMAX(1, s->nmean);
178 /* initialise offset */
179 switch (s->internal_ftype) {
181 s->avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
186 s->avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
189 av_log(s->avctx, AV_LOG_ERROR, "unknown audio type\n");
190 return AVERROR_PATCHWELCOME;
193 for (chan = 0; chan < s->channels; chan++)
194 for (i = 0; i < nblock; i++)
195 s->offset[chan][i] = mean;
199 static int decode_wave_header(AVCodecContext *avctx, const uint8_t *header,
206 bytestream2_init(&gb, header, header_size);
208 if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
209 av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
210 return AVERROR_INVALIDDATA;
213 bytestream2_skip(&gb, 4); /* chunk size */
215 if (bytestream2_get_le32(&gb) != MKTAG('W', 'A', 'V', 'E')) {
216 av_log(avctx, AV_LOG_ERROR, "missing WAVE tag\n");
217 return AVERROR_INVALIDDATA;
220 while (bytestream2_get_le32(&gb) != MKTAG('f', 'm', 't', ' ')) {
221 len = bytestream2_get_le32(&gb);
222 bytestream2_skip(&gb, len);
223 if (len < 0 || bytestream2_get_bytes_left(&gb) < 16) {
224 av_log(avctx, AV_LOG_ERROR, "no fmt chunk found\n");
225 return AVERROR_INVALIDDATA;
228 len = bytestream2_get_le32(&gb);
231 av_log(avctx, AV_LOG_ERROR, "fmt chunk was too short\n");
232 return AVERROR_INVALIDDATA;
235 wave_format = bytestream2_get_le16(&gb);
237 switch (wave_format) {
238 case WAVE_FORMAT_PCM:
241 av_log(avctx, AV_LOG_ERROR, "unsupported wave format\n");
242 return AVERROR(ENOSYS);
245 bytestream2_skip(&gb, 2); // skip channels (already got from shorten header)
246 avctx->sample_rate = bytestream2_get_le32(&gb);
247 bytestream2_skip(&gb, 4); // skip bit rate (represents original uncompressed bit rate)
248 bytestream2_skip(&gb, 2); // skip block align (not needed)
249 bps = bytestream2_get_le16(&gb);
250 avctx->bits_per_coded_sample = bps;
252 if (bps != 16 && bps != 8) {
253 av_log(avctx, AV_LOG_ERROR, "unsupported number of bits per sample: %d\n", bps);
254 return AVERROR(ENOSYS);
259 av_log(avctx, AV_LOG_INFO, "%d header bytes unparsed\n", len);
264 static const int fixed_coeffs[][3] = {
271 static int decode_subframe_lpc(ShortenContext *s, int command, int channel,
272 int residual_size, int32_t coffset)
274 int pred_order, sum, qshift, init_sum, i, j;
277 if (command == FN_QLPC) {
278 /* read/validate prediction order */
279 pred_order = get_ur_golomb_shorten(&s->gb, LPCQSIZE);
280 if ((unsigned)pred_order > s->nwrap) {
281 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
283 return AVERROR(EINVAL);
285 /* read LPC coefficients */
286 for (i = 0; i < pred_order; i++)
287 s->coeffs[i] = get_sr_golomb_shorten(&s->gb, LPCQUANT);
292 /* fixed LPC coeffs */
293 pred_order = command;
294 if (pred_order >= FF_ARRAY_ELEMS(fixed_coeffs)) {
295 av_log(s->avctx, AV_LOG_ERROR, "invalid pred_order %d\n",
297 return AVERROR_INVALIDDATA;
299 coeffs = fixed_coeffs[pred_order];
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) +
318 /* add offset to current samples */
319 if (command == FN_QLPC && coffset)
320 for (i = 0; i < s->blocksize; i++)
321 s->decoded[channel][i] += coffset;
326 static int read_header(ShortenContext *s)
330 /* shorten signature */
331 if (get_bits_long(&s->gb, 32) != AV_RB32("ajkg")) {
332 av_log(s->avctx, AV_LOG_ERROR, "missing shorten magic 'ajkg'\n");
333 return AVERROR_INVALIDDATA;
337 s->blocksize = DEFAULT_BLOCK_SIZE;
339 s->version = get_bits(&s->gb, 8);
340 s->internal_ftype = get_uint(s, TYPESIZE);
342 s->channels = get_uint(s, CHANSIZE);
344 av_log(s->avctx, AV_LOG_ERROR, "No channels reported\n");
345 return AVERROR_INVALIDDATA;
347 if (s->channels > MAX_CHANNELS) {
348 av_log(s->avctx, AV_LOG_ERROR, "too many channels: %d\n", s->channels);
350 return AVERROR_INVALIDDATA;
352 s->avctx->channels = s->channels;
354 /* get blocksize if version > 0 */
355 if (s->version > 0) {
359 blocksize = get_uint(s, av_log2(DEFAULT_BLOCK_SIZE));
360 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
361 av_log(s->avctx, AV_LOG_ERROR,
362 "invalid or unsupported block size: %d\n",
364 return AVERROR(EINVAL);
366 s->blocksize = blocksize;
368 maxnlpc = get_uint(s, LPCQSIZE);
369 s->nmean = get_uint(s, 0);
371 skip_bytes = get_uint(s, NSKIPSIZE);
372 if ((unsigned)skip_bytes > get_bits_left(&s->gb)/8) {
373 av_log(s->avctx, AV_LOG_ERROR, "invalid skip_bytes: %d\n", skip_bytes);
374 return AVERROR_INVALIDDATA;
377 for (i = 0; i < skip_bytes; i++)
378 skip_bits(&s->gb, 8);
380 s->nwrap = FFMAX(NWRAP, maxnlpc);
382 if ((ret = allocate_buffers(s)) < 0)
385 if ((ret = init_offset(s)) < 0)
389 s->lpcqoffset = V2LPCQOFFSET;
391 if (get_ur_golomb_shorten(&s->gb, FNSIZE) != FN_VERBATIM) {
392 av_log(s->avctx, AV_LOG_ERROR,
393 "missing verbatim section at beginning of stream\n");
394 return AVERROR_INVALIDDATA;
397 s->header_size = get_ur_golomb_shorten(&s->gb, VERBATIM_CKSIZE_SIZE);
398 if (s->header_size >= OUT_BUFFER_SIZE ||
399 s->header_size < CANONICAL_HEADER_SIZE) {
400 av_log(s->avctx, AV_LOG_ERROR, "header is wrong size: %d\n",
402 return AVERROR_INVALIDDATA;
405 for (i = 0; i < s->header_size; i++)
406 s->header[i] = (char)get_ur_golomb_shorten(&s->gb, VERBATIM_BYTE_SIZE);
408 if ((ret = decode_wave_header(s->avctx, s->header, s->header_size)) < 0)
419 static int shorten_decode_frame(AVCodecContext *avctx, void *data,
420 int *got_frame_ptr, AVPacket *avpkt)
422 AVFrame *frame = data;
423 const uint8_t *buf = avpkt->data;
424 int buf_size = avpkt->size;
425 ShortenContext *s = avctx->priv_data;
426 int i, input_buf_size = 0;
429 /* allocate internal bitstream buffer */
430 if (s->max_framesize == 0) {
432 s->max_framesize = 8192; // should hopefully be enough for the first header
433 tmp_ptr = av_fast_realloc(s->bitstream, &s->allocated_bitstream_size,
434 s->max_framesize + FF_INPUT_BUFFER_PADDING_SIZE);
436 av_log(avctx, AV_LOG_ERROR, "error allocating bitstream buffer\n");
437 return AVERROR(ENOMEM);
439 memset(tmp_ptr, 0, s->allocated_bitstream_size);
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 + FF_INPUT_BUFFER_PADDING_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 unsigned bitshift = get_ur_golomb_shorten(&s->gb, BITSHIFTSIZE);
515 av_log(avctx, AV_LOG_ERROR, "bitshift %d is invalid\n",
517 return AVERROR_INVALIDDATA;
519 s->bitshift = bitshift;
523 unsigned blocksize = get_uint(s, av_log2(s->blocksize));
524 if (blocksize > s->blocksize) {
525 av_log(avctx, AV_LOG_ERROR,
526 "Increasing block size is not supported\n");
527 return AVERROR_PATCHWELCOME;
529 if (!blocksize || blocksize > MAX_BLOCKSIZE) {
530 av_log(avctx, AV_LOG_ERROR, "invalid or unsupported "
531 "block size: %d\n", blocksize);
532 return AVERROR(EINVAL);
534 s->blocksize = blocksize;
538 s->got_quit_command = 1;
541 if (cmd == FN_BLOCKSIZE || cmd == FN_QUIT) {
546 /* process audio command */
547 int residual_size = 0;
548 int channel = s->cur_chan;
551 /* get Rice code for residual decoding */
552 if (cmd != FN_ZERO) {
553 residual_size = get_ur_golomb_shorten(&s->gb, ENERGYSIZE);
554 /* This is a hack as version 0 differed in the definition
555 * of get_sr_golomb_shorten(). */
560 /* calculate sample offset using means from previous blocks */
562 coffset = s->offset[channel][0];
564 int32_t sum = (s->version < 2) ? 0 : s->nmean / 2;
565 for (i = 0; i < s->nmean; i++)
566 sum += s->offset[channel][i];
567 coffset = sum / s->nmean;
569 coffset = s->bitshift == 0 ? coffset : coffset >> s->bitshift - 1 >> 1;
572 /* decode samples for this channel */
573 if (cmd == FN_ZERO) {
574 for (i = 0; i < s->blocksize; i++)
575 s->decoded[channel][i] = 0;
577 if ((ret = decode_subframe_lpc(s, cmd, channel,
578 residual_size, coffset)) < 0)
582 /* update means with info from the current block */
584 int32_t sum = (s->version < 2) ? 0 : s->blocksize / 2;
585 for (i = 0; i < s->blocksize; i++)
586 sum += s->decoded[channel][i];
588 for (i = 1; i < s->nmean; i++)
589 s->offset[channel][i - 1] = s->offset[channel][i];
592 s->offset[channel][s->nmean - 1] = sum / s->blocksize;
594 s->offset[channel][s->nmean - 1] = (sum / s->blocksize) << s->bitshift;
597 /* copy wrap samples for use with next block */
598 for (i = -s->nwrap; i < 0; i++)
599 s->decoded[channel][i] = s->decoded[channel][i + s->blocksize];
601 /* shift samples to add in unused zero bits which were removed
603 fix_bitshift(s, s->decoded[channel]);
605 /* if this is the last channel in the block, output the samples */
607 if (s->cur_chan == s->channels) {
609 int16_t *samples_s16;
612 /* get output buffer */
613 frame->nb_samples = s->blocksize;
614 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
617 for (chan = 0; chan < s->channels; chan++) {
618 samples_u8 = ((uint8_t **)frame->extended_data)[chan];
619 samples_s16 = ((int16_t **)frame->extended_data)[chan];
620 for (i = 0; i < s->blocksize; i++) {
621 switch (s->internal_ftype) {
623 *samples_u8++ = av_clip_uint8(s->decoded[chan][i]);
627 *samples_s16++ = av_clip_int16(s->decoded[chan][i]);
637 if (s->cur_chan < s->channels)
641 s->bitindex = get_bits_count(&s->gb) - 8 * (get_bits_count(&s->gb) / 8);
642 i = get_bits_count(&s->gb) / 8;
644 av_log(s->avctx, AV_LOG_ERROR, "overread: %d\n", i - buf_size);
645 s->bitstream_size = 0;
646 s->bitstream_index = 0;
647 return AVERROR_INVALIDDATA;
649 if (s->bitstream_size) {
650 s->bitstream_index += i;
651 s->bitstream_size -= i;
652 return input_buf_size;
657 static av_cold int shorten_decode_close(AVCodecContext *avctx)
659 ShortenContext *s = avctx->priv_data;
662 for (i = 0; i < s->channels; i++) {
663 s->decoded[i] = NULL;
664 av_freep(&s->decoded_base[i]);
665 av_freep(&s->offset[i]);
667 av_freep(&s->bitstream);
668 av_freep(&s->coeffs);
673 AVCodec ff_shorten_decoder = {
675 .long_name = NULL_IF_CONFIG_SMALL("Shorten"),
676 .type = AVMEDIA_TYPE_AUDIO,
677 .id = AV_CODEC_ID_SHORTEN,
678 .priv_data_size = sizeof(ShortenContext),
679 .init = shorten_decode_init,
680 .close = shorten_decode_close,
681 .decode = shorten_decode_frame,
682 .capabilities = CODEC_CAP_DELAY | CODEC_CAP_DR1,
683 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
685 AV_SAMPLE_FMT_NONE },