2001-07-22 17:18:56 +03:00
|
|
|
/*
|
|
|
|
* Common bit i/o utils
|
2009-01-19 17:46:40 +02:00
|
|
|
* Copyright (c) 2000, 2001 Fabrice Bellard
|
2004-01-10 18:04:55 +02:00
|
|
|
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
|
2010-03-29 05:50:23 +03:00
|
|
|
* Copyright (c) 2010 Loren Merritt
|
2001-07-22 17:18:56 +03:00
|
|
|
*
|
2007-07-05 13:37:29 +03:00
|
|
|
* alternative bitstream reader & writer by Michael Niedermayer <michaelni@gmx.at>
|
|
|
|
*
|
2006-10-07 18:30:46 +03:00
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2002-05-26 01:45:33 +03:00
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
2006-10-07 18:30:46 +03:00
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
2001-07-22 17:18:56 +03:00
|
|
|
*
|
2006-10-07 18:30:46 +03:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2001-07-22 17:18:56 +03:00
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
2002-05-26 01:45:33 +03:00
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
2001-07-22 17:18:56 +03:00
|
|
|
*
|
2002-05-26 01:45:33 +03:00
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
2006-10-07 18:30:46 +03:00
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
2006-01-13 00:43:26 +02:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
2001-07-22 17:18:56 +03:00
|
|
|
*/
|
2003-03-06 13:32:04 +02:00
|
|
|
|
|
|
|
/**
|
2010-04-20 17:45:34 +03:00
|
|
|
* @file
|
2004-12-29 19:50:25 +02:00
|
|
|
* bitstream api.
|
2003-03-06 13:32:04 +02:00
|
|
|
*/
|
2005-12-17 20:14:38 +02:00
|
|
|
|
2021-06-14 18:03:55 +02:00
|
|
|
#include <inttypes.h>
|
|
|
|
#include <stdint.h>
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <string.h>
|
|
|
|
|
|
|
|
#include "config.h"
|
2012-07-23 23:17:52 +03:00
|
|
|
#include "libavutil/avassert.h"
|
2021-06-14 18:03:55 +02:00
|
|
|
#include "libavutil/bswap.h"
|
|
|
|
#include "libavutil/common.h"
|
|
|
|
#include "libavutil/error.h"
|
|
|
|
#include "libavutil/internal.h"
|
|
|
|
#include "libavutil/intreadwrite.h"
|
|
|
|
#include "libavutil/log.h"
|
|
|
|
#include "libavutil/mem.h"
|
2015-10-18 03:22:59 +02:00
|
|
|
#include "libavutil/qsort.h"
|
2012-10-11 19:50:30 +03:00
|
|
|
#include "mathops.h"
|
2009-04-12 11:35:26 +03:00
|
|
|
#include "put_bits.h"
|
2016-05-14 11:45:01 +02:00
|
|
|
#include "vlc.h"
|
2003-01-27 22:39:29 +02:00
|
|
|
|
2011-04-01 13:46:36 +03:00
|
|
|
const uint8_t ff_log2_run[41]={
|
2008-06-26 19:39:21 +03:00
|
|
|
0, 0, 0, 0, 1, 1, 1, 1,
|
|
|
|
2, 2, 2, 2, 3, 3, 3, 3,
|
|
|
|
4, 4, 5, 5, 6, 6, 7, 7,
|
2011-04-01 13:46:36 +03:00
|
|
|
8, 9,10,11,12,13,14,15,
|
|
|
|
16,17,18,19,20,21,22,23,
|
|
|
|
24,
|
2008-06-26 19:39:21 +03:00
|
|
|
};
|
|
|
|
|
2020-10-26 14:41:39 +02:00
|
|
|
void ff_put_string(PutBitContext *pb, const char *string, int terminate_string)
|
2002-05-04 02:13:24 +03:00
|
|
|
{
|
2013-06-15 11:19:51 +03:00
|
|
|
while (*string) {
|
2009-11-30 01:01:29 +02:00
|
|
|
put_bits(pb, 8, *string);
|
|
|
|
string++;
|
2002-05-04 02:13:24 +03:00
|
|
|
}
|
2013-06-15 11:19:51 +03:00
|
|
|
if (terminate_string)
|
2009-11-30 01:01:29 +02:00
|
|
|
put_bits(pb, 8, 0);
|
2002-05-04 02:13:24 +03:00
|
|
|
}
|
|
|
|
|
2020-10-26 14:41:39 +02:00
|
|
|
void ff_copy_bits(PutBitContext *pb, const uint8_t *src, int length)
|
2007-07-06 17:13:25 +03:00
|
|
|
{
|
2013-06-15 11:19:51 +03:00
|
|
|
int words = length >> 4;
|
|
|
|
int bits = length & 15;
|
2007-07-06 17:13:25 +03:00
|
|
|
int i;
|
|
|
|
|
2013-06-15 11:19:51 +03:00
|
|
|
if (length == 0)
|
|
|
|
return;
|
2007-07-06 17:13:25 +03:00
|
|
|
|
2015-05-25 03:48:45 +02:00
|
|
|
av_assert0(length <= put_bits_left(pb));
|
|
|
|
|
2013-06-15 11:19:51 +03:00
|
|
|
if (CONFIG_SMALL || words < 16 || put_bits_count(pb) & 7) {
|
|
|
|
for (i = 0; i < words; i++)
|
|
|
|
put_bits(pb, 16, AV_RB16(src + 2 * i));
|
|
|
|
} else {
|
|
|
|
for (i = 0; put_bits_count(pb) & 31; i++)
|
2007-07-06 17:13:25 +03:00
|
|
|
put_bits(pb, 8, src[i]);
|
|
|
|
flush_put_bits(pb);
|
2013-06-15 11:19:51 +03:00
|
|
|
memcpy(put_bits_ptr(pb), src + i, 2 * words - i);
|
|
|
|
skip_put_bytes(pb, 2 * words - i);
|
2007-07-06 17:13:25 +03:00
|
|
|
}
|
|
|
|
|
2013-06-15 11:19:51 +03:00
|
|
|
put_bits(pb, bits, AV_RB16(src + 2 * words) >> (16 - bits));
|
2007-07-06 17:13:25 +03:00
|
|
|
}
|
|
|
|
|
2001-07-22 17:18:56 +03:00
|
|
|
/* VLC decoding */
|
|
|
|
|
2013-06-15 11:19:51 +03:00
|
|
|
#define GET_DATA(v, table, i, wrap, size) \
|
|
|
|
{ \
|
|
|
|
const uint8_t *ptr = (const uint8_t *)table + i * wrap; \
|
|
|
|
switch(size) { \
|
|
|
|
case 1: \
|
|
|
|
v = *(const uint8_t *)ptr; \
|
|
|
|
break; \
|
|
|
|
case 2: \
|
|
|
|
v = *(const uint16_t *)ptr; \
|
|
|
|
break; \
|
2017-01-05 12:56:00 +02:00
|
|
|
case 4: \
|
2021-02-21 02:44:18 +02:00
|
|
|
default: \
|
|
|
|
av_assert1(size == 4); \
|
2013-06-15 11:19:51 +03:00
|
|
|
v = *(const uint32_t *)ptr; \
|
|
|
|
break; \
|
|
|
|
} \
|
2001-07-22 17:18:56 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2004-11-27 20:10:06 +02:00
|
|
|
static int alloc_table(VLC *vlc, int size, int use_static)
|
2001-07-22 17:18:56 +03:00
|
|
|
{
|
2013-06-15 11:19:51 +03:00
|
|
|
int index = vlc->table_size;
|
|
|
|
|
2001-07-22 17:18:56 +03:00
|
|
|
vlc->table_size += size;
|
|
|
|
if (vlc->table_size > vlc->table_allocated) {
|
2013-06-15 11:19:51 +03:00
|
|
|
if (use_static)
|
2011-10-05 15:12:42 +03:00
|
|
|
abort(); // cannot do anything, init_vlc() is used with too little memory
|
2001-07-22 17:18:56 +03:00
|
|
|
vlc->table_allocated += (1 << vlc->bits);
|
2013-06-17 00:18:57 +03:00
|
|
|
vlc->table = av_realloc_f(vlc->table, vlc->table_allocated, sizeof(VLC_TYPE) * 2);
|
2013-12-09 22:31:29 +03:00
|
|
|
if (!vlc->table) {
|
2013-12-06 15:44:17 +03:00
|
|
|
vlc->table_allocated = 0;
|
|
|
|
vlc->table_size = 0;
|
2013-06-15 11:28:30 +03:00
|
|
|
return AVERROR(ENOMEM);
|
2013-12-06 15:44:17 +03:00
|
|
|
}
|
2014-06-16 19:45:18 +03:00
|
|
|
memset(vlc->table + vlc->table_allocated - (1 << vlc->bits), 0, sizeof(VLC_TYPE) * 2 << vlc->bits);
|
2001-07-22 17:18:56 +03:00
|
|
|
}
|
|
|
|
return index;
|
|
|
|
}
|
|
|
|
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
#define LOCALBUF_ELEMS 1500 // the maximum currently needed is 1296 by rv34
|
|
|
|
|
2014-09-22 12:01:31 +03:00
|
|
|
typedef struct VLCcode {
|
2010-03-29 05:50:23 +03:00
|
|
|
uint8_t bits;
|
2020-10-26 01:05:26 +02:00
|
|
|
VLC_TYPE symbol;
|
2010-03-29 05:50:23 +03:00
|
|
|
/** codeword, with the first bit-to-be-read in the msb
|
|
|
|
* (even if intended for a little-endian bitstream reader) */
|
|
|
|
uint32_t code;
|
|
|
|
} VLCcode;
|
|
|
|
|
2022-01-04 08:43:59 +02:00
|
|
|
static int vlc_common_init(VLC *vlc, int nb_bits, int nb_codes,
|
|
|
|
VLCcode **buf, int flags)
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
{
|
2022-01-04 08:43:59 +02:00
|
|
|
vlc->bits = nb_bits;
|
|
|
|
vlc->table_size = 0;
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
if (flags & INIT_VLC_USE_NEW_STATIC) {
|
|
|
|
av_assert0(nb_codes <= LOCALBUF_ELEMS);
|
|
|
|
} else {
|
2022-01-04 08:43:59 +02:00
|
|
|
vlc->table = NULL;
|
|
|
|
vlc->table_allocated = 0;
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
}
|
|
|
|
if (nb_codes > LOCALBUF_ELEMS) {
|
|
|
|
*buf = av_malloc_array(nb_codes, sizeof(VLCcode));
|
|
|
|
if (!*buf)
|
|
|
|
return AVERROR(ENOMEM);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-03-29 05:50:23 +03:00
|
|
|
static int compare_vlcspec(const void *a, const void *b)
|
2001-07-22 17:18:56 +03:00
|
|
|
{
|
2013-06-15 11:19:51 +03:00
|
|
|
const VLCcode *sa = a, *sb = b;
|
2010-03-29 05:50:23 +03:00
|
|
|
return (sa->code >> 1) - (sb->code >> 1);
|
|
|
|
}
|
|
|
|
/**
|
|
|
|
* Build VLC decoding tables suitable for use with get_vlc().
|
|
|
|
*
|
2016-04-27 19:45:23 +02:00
|
|
|
* @param vlc the context to be initialized
|
2010-03-29 05:50:23 +03:00
|
|
|
*
|
|
|
|
* @param table_nb_bits max length of vlc codes to store directly in this table
|
|
|
|
* (Longer codes are delegated to subtables.)
|
|
|
|
*
|
|
|
|
* @param nb_codes number of elements in codes[]
|
|
|
|
*
|
|
|
|
* @param codes descriptions of the vlc codes
|
|
|
|
* These must be ordered such that codes going into the same subtable are contiguous.
|
|
|
|
* Sorting by VLCcode.code is sufficient, though not necessary.
|
|
|
|
*/
|
|
|
|
static int build_table(VLC *vlc, int table_nb_bits, int nb_codes,
|
|
|
|
VLCcode *codes, int flags)
|
|
|
|
{
|
|
|
|
int table_size, table_index, index, code_prefix, symbol, subtable_bits;
|
|
|
|
int i, j, k, n, nb, inc;
|
2022-01-04 08:43:59 +02:00
|
|
|
VLC_TYPE (*table)[2];
|
2003-02-11 18:35:48 +02:00
|
|
|
uint32_t code;
|
2001-07-22 17:18:56 +03:00
|
|
|
|
2012-03-23 13:38:20 +03:00
|
|
|
if (table_nb_bits > 30)
|
2018-12-31 19:21:41 +02:00
|
|
|
return AVERROR(EINVAL);
|
2020-06-25 15:10:35 +02:00
|
|
|
table_size = 1 << table_nb_bits;
|
2009-09-24 18:13:34 +03:00
|
|
|
table_index = alloc_table(vlc, table_size, flags & INIT_VLC_USE_NEW_STATIC);
|
2015-03-16 10:57:36 +02:00
|
|
|
ff_dlog(NULL, "new table index=%d size=%d\n", table_index, table_size);
|
2001-07-22 17:18:56 +03:00
|
|
|
if (table_index < 0)
|
2013-06-15 11:28:30 +03:00
|
|
|
return table_index;
|
2022-01-04 08:43:59 +02:00
|
|
|
table = &vlc->table[table_index];
|
2001-07-22 17:18:56 +03:00
|
|
|
|
2012-12-19 20:48:21 +03:00
|
|
|
/* first pass: map codes and compute auxiliary table sizes */
|
2010-03-29 05:50:41 +03:00
|
|
|
for (i = 0; i < nb_codes; i++) {
|
2013-06-15 11:19:51 +03:00
|
|
|
n = codes[i].bits;
|
|
|
|
code = codes[i].code;
|
2010-03-29 05:50:23 +03:00
|
|
|
symbol = codes[i].symbol;
|
2017-03-27 21:31:46 +02:00
|
|
|
ff_dlog(NULL, "i=%d n=%d code=0x%"PRIx32"\n", i, n, code);
|
2010-03-29 05:50:41 +03:00
|
|
|
if (n <= table_nb_bits) {
|
|
|
|
/* no need to add another table */
|
|
|
|
j = code >> (32 - table_nb_bits);
|
|
|
|
nb = 1 << (table_nb_bits - n);
|
|
|
|
inc = 1;
|
2020-10-12 05:24:42 +02:00
|
|
|
if (flags & INIT_VLC_OUTPUT_LE) {
|
2010-03-29 05:50:41 +03:00
|
|
|
j = bitswap_32(code);
|
|
|
|
inc = 1 << n;
|
|
|
|
}
|
|
|
|
for (k = 0; k < nb; k++) {
|
2014-06-16 20:40:02 +03:00
|
|
|
int bits = table[j][1];
|
2019-06-05 12:18:54 +02:00
|
|
|
int oldsym = table[j][0];
|
2015-03-16 10:57:36 +02:00
|
|
|
ff_dlog(NULL, "%4x: code=%d n=%d\n", j, i, n);
|
2019-06-05 12:18:54 +02:00
|
|
|
if ((bits || oldsym) && (bits != n || oldsym != symbol)) {
|
2010-03-29 05:50:41 +03:00
|
|
|
av_log(NULL, AV_LOG_ERROR, "incorrect codes\n");
|
2013-06-15 11:28:30 +03:00
|
|
|
return AVERROR_INVALIDDATA;
|
2010-03-29 05:50:41 +03:00
|
|
|
}
|
|
|
|
table[j][1] = n; //bits
|
|
|
|
table[j][0] = symbol;
|
|
|
|
j += inc;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* fill auxiliary table recursively */
|
|
|
|
n -= table_nb_bits;
|
|
|
|
code_prefix = code >> (32 - table_nb_bits);
|
|
|
|
subtable_bits = n;
|
|
|
|
codes[i].bits = n;
|
|
|
|
codes[i].code = code << table_nb_bits;
|
|
|
|
for (k = i+1; k < nb_codes; k++) {
|
|
|
|
n = codes[k].bits - table_nb_bits;
|
|
|
|
if (n <= 0)
|
|
|
|
break;
|
|
|
|
code = codes[k].code;
|
|
|
|
if (code >> (32 - table_nb_bits) != code_prefix)
|
|
|
|
break;
|
|
|
|
codes[k].bits = n;
|
|
|
|
codes[k].code = code << table_nb_bits;
|
|
|
|
subtable_bits = FFMAX(subtable_bits, n);
|
|
|
|
}
|
|
|
|
subtable_bits = FFMIN(subtable_bits, table_nb_bits);
|
2020-10-12 05:24:42 +02:00
|
|
|
j = (flags & INIT_VLC_OUTPUT_LE) ? bitswap_32(code_prefix) >> (32 - table_nb_bits) : code_prefix;
|
2010-03-29 05:50:41 +03:00
|
|
|
table[j][1] = -subtable_bits;
|
2015-03-16 10:57:36 +02:00
|
|
|
ff_dlog(NULL, "%4x: n=%d (subtable)\n",
|
2011-06-06 02:25:32 +03:00
|
|
|
j, codes[i].bits + table_nb_bits);
|
2010-03-29 05:50:41 +03:00
|
|
|
index = build_table(vlc, subtable_bits, k-i, codes+i, flags);
|
|
|
|
if (index < 0)
|
2013-06-15 11:28:30 +03:00
|
|
|
return index;
|
2010-03-29 05:50:41 +03:00
|
|
|
/* note: realloc has been done, so reload tables */
|
2022-01-04 08:43:59 +02:00
|
|
|
table = &vlc->table[table_index];
|
2010-03-29 05:50:41 +03:00
|
|
|
table[j][0] = index; //code
|
2019-06-05 12:18:54 +02:00
|
|
|
if (table[j][0] != index) {
|
|
|
|
avpriv_request_sample(NULL, "strange codes");
|
|
|
|
return AVERROR_PATCHWELCOME;
|
|
|
|
}
|
2010-03-29 05:50:41 +03:00
|
|
|
i = k-1;
|
|
|
|
}
|
2001-07-22 17:18:56 +03:00
|
|
|
}
|
2014-06-16 19:46:14 +03:00
|
|
|
|
|
|
|
for (i = 0; i < table_size; i++) {
|
|
|
|
if (table[i][1] == 0) //bits
|
|
|
|
table[i][0] = -1; //codes
|
|
|
|
}
|
|
|
|
|
2001-07-22 17:18:56 +03:00
|
|
|
return table_index;
|
|
|
|
}
|
|
|
|
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
static int vlc_common_end(VLC *vlc, int nb_bits, int nb_codes, VLCcode *codes,
|
2022-01-04 08:43:59 +02:00
|
|
|
int flags, VLCcode localbuf[LOCALBUF_ELEMS])
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
{
|
|
|
|
int ret = build_table(vlc, nb_bits, nb_codes, codes, flags);
|
|
|
|
|
|
|
|
if (flags & INIT_VLC_USE_NEW_STATIC) {
|
2020-11-03 20:20:15 +02:00
|
|
|
if (vlc->table_size != vlc->table_allocated &&
|
|
|
|
!(flags & (INIT_VLC_STATIC_OVERLONG & ~INIT_VLC_USE_NEW_STATIC)))
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
av_log(NULL, AV_LOG_ERROR, "needed %d had %d\n", vlc->table_size, vlc->table_allocated);
|
|
|
|
av_assert0(ret >= 0);
|
|
|
|
} else {
|
|
|
|
if (codes != localbuf)
|
|
|
|
av_free(codes);
|
|
|
|
if (ret < 0) {
|
|
|
|
av_freep(&vlc->table);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
2001-07-22 17:18:56 +03:00
|
|
|
|
2001-08-06 03:47:03 +03:00
|
|
|
/* Build VLC decoding tables suitable for use with get_vlc().
|
|
|
|
|
2016-04-27 19:45:23 +02:00
|
|
|
'nb_bits' sets the decoding table size (2^nb_bits) entries. The
|
2001-08-06 03:47:03 +03:00
|
|
|
bigger it is, the faster is the decoding. But it should not be too
|
|
|
|
big to save memory and L1 cache. '9' is a good compromise.
|
2005-12-17 20:14:38 +02:00
|
|
|
|
2001-08-06 03:47:03 +03:00
|
|
|
'nb_codes' : number of vlcs codes
|
|
|
|
|
|
|
|
'bits' : table which gives the size (in bits) of each vlc code.
|
|
|
|
|
|
|
|
'codes' : table which gives the bit pattern of of each vlc code.
|
|
|
|
|
2007-05-24 20:38:56 +03:00
|
|
|
'symbols' : table which gives the values to be returned from get_vlc().
|
|
|
|
|
2001-08-06 03:47:03 +03:00
|
|
|
'xxx_wrap' : give the number of bytes between each entry of the
|
|
|
|
'bits' or 'codes' tables.
|
|
|
|
|
|
|
|
'xxx_size' : gives the number of bytes of each entry of the 'bits'
|
2017-01-05 12:57:17 +02:00
|
|
|
or 'codes' tables. Currently 1,2 and 4 are supported.
|
2001-08-06 03:47:03 +03:00
|
|
|
|
2015-06-14 20:28:28 +02:00
|
|
|
'wrap' and 'size' make it possible to use any memory configuration and types
|
2007-05-24 20:38:56 +03:00
|
|
|
(byte/word/long) to store the 'bits', 'codes', and 'symbols' tables.
|
2001-08-06 03:47:03 +03:00
|
|
|
*/
|
2022-01-04 08:43:59 +02:00
|
|
|
int ff_init_vlc_sparse(VLC *vlc, int nb_bits, int nb_codes,
|
2013-06-15 11:19:51 +03:00
|
|
|
const void *bits, int bits_wrap, int bits_size,
|
|
|
|
const void *codes, int codes_wrap, int codes_size,
|
|
|
|
const void *symbols, int symbols_wrap, int symbols_size,
|
|
|
|
int flags)
|
2001-07-22 17:18:56 +03:00
|
|
|
{
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
VLCcode localbuf[LOCALBUF_ELEMS], *buf = localbuf;
|
2010-05-13 01:38:05 +03:00
|
|
|
int i, j, ret;
|
2010-03-29 05:50:23 +03:00
|
|
|
|
2022-01-04 08:43:59 +02:00
|
|
|
ret = vlc_common_init(vlc, nb_bits, nb_codes, &buf, flags);
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
2010-05-13 01:38:05 +03:00
|
|
|
|
2012-07-23 23:17:52 +03:00
|
|
|
av_assert0(symbols_size <= 2 || !symbols);
|
2010-03-29 05:50:23 +03:00
|
|
|
j = 0;
|
|
|
|
#define COPY(condition)\
|
2013-06-15 11:19:51 +03:00
|
|
|
for (i = 0; i < nb_codes; i++) { \
|
2020-10-24 14:11:14 +02:00
|
|
|
unsigned len; \
|
|
|
|
GET_DATA(len, bits, i, bits_wrap, bits_size); \
|
2013-06-15 11:19:51 +03:00
|
|
|
if (!(condition)) \
|
|
|
|
continue; \
|
2020-10-24 14:11:14 +02:00
|
|
|
if (len > 3*nb_bits || len > 32) { \
|
|
|
|
av_log(NULL, AV_LOG_ERROR, "Too long VLC (%u) in init_vlc\n", len);\
|
2020-06-25 13:20:19 +02:00
|
|
|
if (buf != localbuf) \
|
2013-08-08 16:05:48 +03:00
|
|
|
av_free(buf); \
|
2018-12-31 19:21:41 +02:00
|
|
|
return AVERROR(EINVAL); \
|
2013-06-17 00:18:57 +03:00
|
|
|
} \
|
2020-10-24 14:11:14 +02:00
|
|
|
buf[j].bits = len; \
|
2013-06-15 11:19:51 +03:00
|
|
|
GET_DATA(buf[j].code, codes, i, codes_wrap, codes_size); \
|
2013-06-17 00:18:57 +03:00
|
|
|
if (buf[j].code >= (1LL<<buf[j].bits)) { \
|
2017-03-27 21:31:46 +02:00
|
|
|
av_log(NULL, AV_LOG_ERROR, "Invalid code %"PRIx32" for %d in " \
|
|
|
|
"init_vlc\n", buf[j].code, i); \
|
2020-06-25 13:20:19 +02:00
|
|
|
if (buf != localbuf) \
|
2013-08-08 16:05:48 +03:00
|
|
|
av_free(buf); \
|
2018-12-31 19:21:41 +02:00
|
|
|
return AVERROR(EINVAL); \
|
2013-06-17 00:18:57 +03:00
|
|
|
} \
|
2020-10-12 05:24:42 +02:00
|
|
|
if (flags & INIT_VLC_INPUT_LE) \
|
2013-06-15 11:19:51 +03:00
|
|
|
buf[j].code = bitswap_32(buf[j].code); \
|
|
|
|
else \
|
|
|
|
buf[j].code <<= 32 - buf[j].bits; \
|
|
|
|
if (symbols) \
|
|
|
|
GET_DATA(buf[j].symbol, symbols, i, symbols_wrap, symbols_size) \
|
2013-06-17 00:18:57 +03:00
|
|
|
else \
|
|
|
|
buf[j].symbol = i; \
|
2013-06-15 11:19:51 +03:00
|
|
|
j++; \
|
2010-03-29 05:50:23 +03:00
|
|
|
}
|
2020-10-24 14:11:14 +02:00
|
|
|
COPY(len > nb_bits);
|
2010-03-29 05:50:23 +03:00
|
|
|
// qsort is the slowest part of init_vlc, and could probably be improved or avoided
|
2015-10-18 03:22:59 +02:00
|
|
|
AV_QSORT(buf, j, struct VLCcode, compare_vlcspec);
|
2020-10-24 14:11:14 +02:00
|
|
|
COPY(len && len <= nb_bits);
|
2010-03-29 05:50:23 +03:00
|
|
|
nb_codes = j;
|
|
|
|
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
return vlc_common_end(vlc, nb_bits, nb_codes, buf,
|
2022-01-04 08:43:59 +02:00
|
|
|
flags, localbuf);
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
}
|
2014-06-16 20:40:02 +03:00
|
|
|
|
2022-01-04 08:43:59 +02:00
|
|
|
int ff_init_vlc_from_lengths(VLC *vlc, int nb_bits, int nb_codes,
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
const int8_t *lens, int lens_wrap,
|
|
|
|
const void *symbols, int symbols_wrap, int symbols_size,
|
|
|
|
int offset, int flags, void *logctx)
|
|
|
|
{
|
|
|
|
VLCcode localbuf[LOCALBUF_ELEMS], *buf = localbuf;
|
|
|
|
uint64_t code;
|
|
|
|
int ret, j, len_max = FFMIN(32, 3 * nb_bits);
|
|
|
|
|
2022-01-04 08:43:59 +02:00
|
|
|
ret = vlc_common_init(vlc, nb_bits, nb_codes, &buf, flags);
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
j = code = 0;
|
|
|
|
for (int i = 0; i < nb_codes; i++, lens += lens_wrap) {
|
|
|
|
int len = *lens;
|
|
|
|
if (len > 0) {
|
|
|
|
unsigned sym;
|
|
|
|
|
|
|
|
buf[j].bits = len;
|
|
|
|
if (symbols)
|
|
|
|
GET_DATA(sym, symbols, i, symbols_wrap, symbols_size)
|
|
|
|
else
|
|
|
|
sym = i;
|
|
|
|
buf[j].symbol = sym + offset;
|
|
|
|
buf[j++].code = code;
|
|
|
|
} else if (len < 0) {
|
|
|
|
len = -len;
|
|
|
|
} else
|
|
|
|
continue;
|
|
|
|
if (len > len_max || code & ((1U << (32 - len)) - 1)) {
|
|
|
|
av_log(logctx, AV_LOG_ERROR, "Invalid VLC (length %u)\n", len);
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
code += 1U << (32 - len);
|
|
|
|
if (code > UINT32_MAX + 1ULL) {
|
|
|
|
av_log(logctx, AV_LOG_ERROR, "Overdetermined VLC tree\n");
|
|
|
|
goto fail;
|
2013-07-13 14:32:39 +03:00
|
|
|
}
|
2001-07-22 17:18:56 +03:00
|
|
|
}
|
2022-01-04 08:43:59 +02:00
|
|
|
return vlc_common_end(vlc, nb_bits, j, buf, flags, localbuf);
|
avcodec/bitstream: Add second function to create VLCs
When using ff_init_vlc_sparse() to create a VLC, three input tables are
used: A table for lengths, one for codes and one for symbols; the latter
one can be omitted, then a default one will be used. These input tables
will be traversed twice, once to get the long codes (which will be put
into subtables) and once for the small codes. The long codes are then
sorted so that entries that should be in the same subtable are
contiguous.
This commit adds an alternative to ff_init_vlc_sparse():
ff_init_vlc_from_lengths(). It is based upon the observation that if
lengths, codes and symbols tables are permuted (in the same way) so that
the codes are ordered from left to right in the corresponding tree and
if said tree is complete (i.e. every non-leaf node has two children),
the codes can be easily computed from the lengths and are therefore
redundant. This means that if one initializes such a VLC with explicitly
coded lengths, codes and symbols, the codes can be avoided; and even if
one has no explicitly coded symbols, it might still be beneficial to
remove the codes even when one has to add a new symbol table, because
codes are typically longer than symbols so that the latter often fit
into a smaller type, saving space.
Furthermore, given that the codes here are by definition ordered from
left to right, it is unnecessary to sort them again; for the same
reason, one does not have to traverse the input twice. This function
proved to be faster than ff_init_vlc_sparse() whenever it has been
benchmarked.
This function is usable for static tables (they can simply be permuted
once) as well as in scenarios where the tables are naturally ordered
from left to right in the tree; the latter e.g. happens with Smacker,
Theora and several other formats.
In order to make it also usable for (static) tables with incomplete trees,
negative lengths are used to indicate that there is an open end of a
certain length.
Finally, ff_init_vlc_from_lengths() has one downside compared to
ff_init_vlc_sparse(): The latter uses tables that can be reused by
encoders. Of course, one could calculate the needed table at runtime
if one so wishes, but it is nevertheless an obstacle.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-10-25 00:54:21 +02:00
|
|
|
fail:
|
|
|
|
if (buf != localbuf)
|
|
|
|
av_free(buf);
|
|
|
|
return AVERROR_INVALIDDATA;
|
2001-07-22 17:18:56 +03:00
|
|
|
}
|
|
|
|
|
2012-02-09 12:37:58 +03:00
|
|
|
void ff_free_vlc(VLC *vlc)
|
2001-07-22 17:18:56 +03:00
|
|
|
{
|
2007-07-05 09:47:00 +03:00
|
|
|
av_freep(&vlc->table);
|
2001-07-22 17:18:56 +03:00
|
|
|
}
|