|
|
|
/*
|
|
|
|
* copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
|
|
|
|
*
|
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @file
|
|
|
|
* bitstream writer API
|
|
|
|
*/
|
|
|
|
|
|
|
|
#ifndef AVCODEC_PUT_BITS_H
|
|
|
|
#define AVCODEC_PUT_BITS_H
|
|
|
|
|
|
|
|
#include <stdint.h>
|
|
|
|
#include <stddef.h>
|
|
|
|
|
avcodec/put_bits: Make bit buffers 64-bit
Change BitBuf into uint64_t on 64-bit x86. This means we need to flush the
buffer less often, which is a significant speed win. All other platforms,
including all 32-bit ones, are unchanged. Output bitstream is the same.
All API constraints are kept in place, e.g., you still cannot put_bits()
more than 31 bits at a time. This is so that codecs cannot accidentally
become 64-bit-only or similar.
Benchmarking on transcoding to various formats shows consistently
positive results:
dnxhd 25.60 fps -> 26.26 fps ( +2.6%)
dvvideo 24.88 fps -> 25.17 fps ( +1.2%)
ffv1 14.32 fps -> 14.58 fps ( +1.8%)
huffyuv 58.75 fps -> 63.27 fps ( +7.7%)
jpegls 6.22 fps -> 6.34 fps ( +1.8%)
magicyuv 57.10 fps -> 63.29 fps (+10.8%)
mjpeg 48.65 fps -> 49.01 fps ( +0.7%)
mpeg1video 76.41 fps -> 77.01 fps ( +0.8%)
mpeg2video 75.99 fps -> 77.43 fps ( +1.9%)
mpeg4 80.66 fps -> 81.37 fps ( +0.9%)
prores 12.35 fps -> 12.88 fps ( +4.3%)
prores_ks 16.20 fps -> 16.80 fps ( +3.7%)
rv20 62.80 fps -> 62.99 fps ( +0.3%)
utvideo 68.41 fps -> 76.32 fps (+11.6%)
Note that this includes video decoding and all other encoding work,
such as DCTs. If you isolate the actual bit-writing routines, it is
likely to be much more.
Benchmark details: Transcoding the first 30 seconds of Big Buck Bunny
in 1080p, Haswell 2.1 GHz, GCC 8.3, generally quantizer locked to
5.0. (Exceptions: DNxHD needs fixed bitrate, and JPEG-LS is so slow
that I only took the first 10 seconds, not 30.) All runs were done
ten times and single-threaded, top and bottom two results discarded to
get rid of outliers, arithmetic mean between the remaining six.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
4 years ago
|
|
|
#include "config.h"
|
|
|
|
#include "libavutil/intreadwrite.h"
|
|
|
|
#include "libavutil/avassert.h"
|
|
|
|
|
|
|
|
#include "version.h"
|
|
|
|
|
avcodec/put_bits: Make bit buffers 64-bit
Change BitBuf into uint64_t on 64-bit x86. This means we need to flush the
buffer less often, which is a significant speed win. All other platforms,
including all 32-bit ones, are unchanged. Output bitstream is the same.
All API constraints are kept in place, e.g., you still cannot put_bits()
more than 31 bits at a time. This is so that codecs cannot accidentally
become 64-bit-only or similar.
Benchmarking on transcoding to various formats shows consistently
positive results:
dnxhd 25.60 fps -> 26.26 fps ( +2.6%)
dvvideo 24.88 fps -> 25.17 fps ( +1.2%)
ffv1 14.32 fps -> 14.58 fps ( +1.8%)
huffyuv 58.75 fps -> 63.27 fps ( +7.7%)
jpegls 6.22 fps -> 6.34 fps ( +1.8%)
magicyuv 57.10 fps -> 63.29 fps (+10.8%)
mjpeg 48.65 fps -> 49.01 fps ( +0.7%)
mpeg1video 76.41 fps -> 77.01 fps ( +0.8%)
mpeg2video 75.99 fps -> 77.43 fps ( +1.9%)
mpeg4 80.66 fps -> 81.37 fps ( +0.9%)
prores 12.35 fps -> 12.88 fps ( +4.3%)
prores_ks 16.20 fps -> 16.80 fps ( +3.7%)
rv20 62.80 fps -> 62.99 fps ( +0.3%)
utvideo 68.41 fps -> 76.32 fps (+11.6%)
Note that this includes video decoding and all other encoding work,
such as DCTs. If you isolate the actual bit-writing routines, it is
likely to be much more.
Benchmark details: Transcoding the first 30 seconds of Big Buck Bunny
in 1080p, Haswell 2.1 GHz, GCC 8.3, generally quantizer locked to
5.0. (Exceptions: DNxHD needs fixed bitrate, and JPEG-LS is so slow
that I only took the first 10 seconds, not 30.) All runs were done
ten times and single-threaded, top and bottom two results discarded to
get rid of outliers, arithmetic mean between the remaining six.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
4 years ago
|
|
|
#if ARCH_X86_64
|
|
|
|
// TODO: Benchmark and optionally enable on other 64-bit architectures.
|
|
|
|
typedef uint64_t BitBuf;
|
|
|
|
#define AV_WBBUF AV_WB64
|
|
|
|
#define AV_WLBUF AV_WL64
|
|
|
|
#else
|
|
|
|
typedef uint32_t BitBuf;
|
|
|
|
#define AV_WBBUF AV_WB32
|
|
|
|
#define AV_WLBUF AV_WL32
|
avcodec/put_bits: Make bit buffers 64-bit
Change BitBuf into uint64_t on 64-bit x86. This means we need to flush the
buffer less often, which is a significant speed win. All other platforms,
including all 32-bit ones, are unchanged. Output bitstream is the same.
All API constraints are kept in place, e.g., you still cannot put_bits()
more than 31 bits at a time. This is so that codecs cannot accidentally
become 64-bit-only or similar.
Benchmarking on transcoding to various formats shows consistently
positive results:
dnxhd 25.60 fps -> 26.26 fps ( +2.6%)
dvvideo 24.88 fps -> 25.17 fps ( +1.2%)
ffv1 14.32 fps -> 14.58 fps ( +1.8%)
huffyuv 58.75 fps -> 63.27 fps ( +7.7%)
jpegls 6.22 fps -> 6.34 fps ( +1.8%)
magicyuv 57.10 fps -> 63.29 fps (+10.8%)
mjpeg 48.65 fps -> 49.01 fps ( +0.7%)
mpeg1video 76.41 fps -> 77.01 fps ( +0.8%)
mpeg2video 75.99 fps -> 77.43 fps ( +1.9%)
mpeg4 80.66 fps -> 81.37 fps ( +0.9%)
prores 12.35 fps -> 12.88 fps ( +4.3%)
prores_ks 16.20 fps -> 16.80 fps ( +3.7%)
rv20 62.80 fps -> 62.99 fps ( +0.3%)
utvideo 68.41 fps -> 76.32 fps (+11.6%)
Note that this includes video decoding and all other encoding work,
such as DCTs. If you isolate the actual bit-writing routines, it is
likely to be much more.
Benchmark details: Transcoding the first 30 seconds of Big Buck Bunny
in 1080p, Haswell 2.1 GHz, GCC 8.3, generally quantizer locked to
5.0. (Exceptions: DNxHD needs fixed bitrate, and JPEG-LS is so slow
that I only took the first 10 seconds, not 30.) All runs were done
ten times and single-threaded, top and bottom two results discarded to
get rid of outliers, arithmetic mean between the remaining six.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
4 years ago
|
|
|
#endif
|
|
|
|
|
|
|
|
static const int BUF_BITS = 8 * sizeof(BitBuf);
|
|
|
|
|
|
|
|
typedef struct PutBitContext {
|
|
|
|
BitBuf bit_buf;
|
|
|
|
int bit_left;
|
|
|
|
uint8_t *buf, *buf_ptr, *buf_end;
|
|
|
|
#if LIBAVCODEC_VERSION_MAJOR < 59
|
|
|
|
int size_in_bits;
|
|
|
|
#endif
|
|
|
|
} PutBitContext;
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Initialize the PutBitContext s.
|
|
|
|
*
|
|
|
|
* @param buffer the buffer where to put bits
|
|
|
|
* @param buffer_size the size in bytes of buffer
|
|
|
|
*/
|
|
|
|
static inline void init_put_bits(PutBitContext *s, uint8_t *buffer,
|
|
|
|
int buffer_size)
|
|
|
|
{
|
|
|
|
if (buffer_size < 0) {
|
|
|
|
buffer_size = 0;
|
|
|
|
buffer = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
s->buf = buffer;
|
|
|
|
s->buf_end = s->buf + buffer_size;
|
|
|
|
s->buf_ptr = s->buf;
|
|
|
|
s->bit_left = BUF_BITS;
|
|
|
|
s->bit_buf = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @return the total number of bits written to the bitstream.
|
|
|
|
*/
|
|
|
|
static inline int put_bits_count(PutBitContext *s)
|
|
|
|
{
|
|
|
|
return (s->buf_ptr - s->buf) * 8 + BUF_BITS - s->bit_left;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @return the number of bytes output so far; may only be called
|
|
|
|
* when the PutBitContext is freshly initialized or flushed.
|
|
|
|
*/
|
|
|
|
static inline int put_bytes_output(const PutBitContext *s)
|
|
|
|
{
|
|
|
|
av_assert2(s->bit_left == BUF_BITS);
|
|
|
|
return s->buf_ptr - s->buf;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param round_up When set, the number of bits written so far will be
|
|
|
|
* rounded up to the next byte.
|
|
|
|
* @return the number of bytes output so far.
|
|
|
|
*/
|
|
|
|
static inline int put_bytes_count(const PutBitContext *s, int round_up)
|
|
|
|
{
|
|
|
|
return s->buf_ptr - s->buf + ((BUF_BITS - s->bit_left + (round_up ? 7 : 0)) >> 3);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Rebase the bit writer onto a reallocated buffer.
|
|
|
|
*
|
|
|
|
* @param buffer the buffer where to put bits
|
|
|
|
* @param buffer_size the size in bytes of buffer,
|
|
|
|
* must be large enough to hold everything written so far
|
|
|
|
*/
|
|
|
|
static inline void rebase_put_bits(PutBitContext *s, uint8_t *buffer,
|
|
|
|
int buffer_size)
|
|
|
|
{
|
|
|
|
av_assert0(8*buffer_size >= put_bits_count(s));
|
|
|
|
|
|
|
|
s->buf_end = buffer + buffer_size;
|
|
|
|
s->buf_ptr = buffer + (s->buf_ptr - s->buf);
|
|
|
|
s->buf = buffer;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @return the number of bits available in the bitstream.
|
|
|
|
*/
|
|
|
|
static inline int put_bits_left(PutBitContext* s)
|
|
|
|
{
|
|
|
|
return (s->buf_end - s->buf_ptr) * 8 - BUF_BITS + s->bit_left;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @param round_up When set, the number of bits written will be
|
|
|
|
* rounded up to the next byte.
|
|
|
|
* @return the number of bytes left.
|
|
|
|
*/
|
|
|
|
static inline int put_bytes_left(const PutBitContext *s, int round_up)
|
|
|
|
{
|
|
|
|
return s->buf_end - s->buf_ptr - ((BUF_BITS - s->bit_left + (round_up ? 7 : 0)) >> 3);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Pad the end of the output stream with zeros.
|
|
|
|
*/
|
|
|
|
static inline void flush_put_bits(PutBitContext *s)
|
|
|
|
{
|
|
|
|
#ifndef BITSTREAM_WRITER_LE
|
|
|
|
if (s->bit_left < BUF_BITS)
|
|
|
|
s->bit_buf <<= s->bit_left;
|
|
|
|
#endif
|
|
|
|
while (s->bit_left < BUF_BITS) {
|
|
|
|
av_assert0(s->buf_ptr < s->buf_end);
|
|
|
|
#ifdef BITSTREAM_WRITER_LE
|
|
|
|
*s->buf_ptr++ = s->bit_buf;
|
|
|
|
s->bit_buf >>= 8;
|
|
|
|
#else
|
|
|
|
*s->buf_ptr++ = s->bit_buf >> (BUF_BITS - 8);
|
|
|
|
s->bit_buf <<= 8;
|
|
|
|
#endif
|
|
|
|
s->bit_left += 8;
|
|
|
|
}
|
|
|
|
s->bit_left = BUF_BITS;
|
|
|
|
s->bit_buf = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void flush_put_bits_le(PutBitContext *s)
|
|
|
|
{
|
|
|
|
while (s->bit_left < BUF_BITS) {
|
|
|
|
av_assert0(s->buf_ptr < s->buf_end);
|
|
|
|
*s->buf_ptr++ = s->bit_buf;
|
|
|
|
s->bit_buf >>= 8;
|
|
|
|
s->bit_left += 8;
|
|
|
|
}
|
|
|
|
s->bit_left = BUF_BITS;
|
|
|
|
s->bit_buf = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
#if FF_API_AVPRIV_PUT_BITS
|
|
|
|
void avpriv_align_put_bits(PutBitContext *s);
|
|
|
|
void avpriv_copy_bits(PutBitContext *pb, const uint8_t *src, int length);
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef BITSTREAM_WRITER_LE
|
|
|
|
#define ff_put_string ff_put_string_unsupported_here
|
|
|
|
#define ff_copy_bits ff_copy_bits_unsupported_here
|
|
|
|
#else
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Put the string string in the bitstream.
|
|
|
|
*
|
|
|
|
* @param terminate_string 0-terminates the written string if value is 1
|
|
|
|
*/
|
|
|
|
void ff_put_string(PutBitContext *pb, const char *string,
|
|
|
|
int terminate_string);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Copy the content of src to the bitstream.
|
|
|
|
*
|
|
|
|
* @param length the number of bits of src to copy
|
|
|
|
*/
|
|
|
|
void ff_copy_bits(PutBitContext *pb, const uint8_t *src, int length);
|
|
|
|
#endif
|
|
|
|
|
avcodec/put_bits: Make bit buffers 64-bit
Change BitBuf into uint64_t on 64-bit x86. This means we need to flush the
buffer less often, which is a significant speed win. All other platforms,
including all 32-bit ones, are unchanged. Output bitstream is the same.
All API constraints are kept in place, e.g., you still cannot put_bits()
more than 31 bits at a time. This is so that codecs cannot accidentally
become 64-bit-only or similar.
Benchmarking on transcoding to various formats shows consistently
positive results:
dnxhd 25.60 fps -> 26.26 fps ( +2.6%)
dvvideo 24.88 fps -> 25.17 fps ( +1.2%)
ffv1 14.32 fps -> 14.58 fps ( +1.8%)
huffyuv 58.75 fps -> 63.27 fps ( +7.7%)
jpegls 6.22 fps -> 6.34 fps ( +1.8%)
magicyuv 57.10 fps -> 63.29 fps (+10.8%)
mjpeg 48.65 fps -> 49.01 fps ( +0.7%)
mpeg1video 76.41 fps -> 77.01 fps ( +0.8%)
mpeg2video 75.99 fps -> 77.43 fps ( +1.9%)
mpeg4 80.66 fps -> 81.37 fps ( +0.9%)
prores 12.35 fps -> 12.88 fps ( +4.3%)
prores_ks 16.20 fps -> 16.80 fps ( +3.7%)
rv20 62.80 fps -> 62.99 fps ( +0.3%)
utvideo 68.41 fps -> 76.32 fps (+11.6%)
Note that this includes video decoding and all other encoding work,
such as DCTs. If you isolate the actual bit-writing routines, it is
likely to be much more.
Benchmark details: Transcoding the first 30 seconds of Big Buck Bunny
in 1080p, Haswell 2.1 GHz, GCC 8.3, generally quantizer locked to
5.0. (Exceptions: DNxHD needs fixed bitrate, and JPEG-LS is so slow
that I only took the first 10 seconds, not 30.) All runs were done
ten times and single-threaded, top and bottom two results discarded to
get rid of outliers, arithmetic mean between the remaining six.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
4 years ago
|
|
|
static inline void put_bits_no_assert(PutBitContext *s, int n, BitBuf value)
|
|
|
|
{
|
|
|
|
BitBuf bit_buf;
|
|
|
|
int bit_left;
|
|
|
|
|
|
|
|
bit_buf = s->bit_buf;
|
|
|
|
bit_left = s->bit_left;
|
|
|
|
|
|
|
|
/* XXX: optimize */
|
|
|
|
#ifdef BITSTREAM_WRITER_LE
|
|
|
|
bit_buf |= value << (BUF_BITS - bit_left);
|
|
|
|
if (n >= bit_left) {
|
|
|
|
if (s->buf_end - s->buf_ptr >= sizeof(BitBuf)) {
|
|
|
|
AV_WLBUF(s->buf_ptr, bit_buf);
|
|
|
|
s->buf_ptr += sizeof(BitBuf);
|
|
|
|
} else {
|
|
|
|
av_log(NULL, AV_LOG_ERROR, "Internal error, put_bits buffer too small\n");
|
|
|
|
av_assert2(0);
|
|
|
|
}
|
|
|
|
bit_buf = value >> bit_left;
|
|
|
|
bit_left += BUF_BITS;
|
|
|
|
}
|
|
|
|
bit_left -= n;
|
|
|
|
#else
|
|
|
|
if (n < bit_left) {
|
|
|
|
bit_buf = (bit_buf << n) | value;
|
|
|
|
bit_left -= n;
|
|
|
|
} else {
|
|
|
|
bit_buf <<= bit_left;
|
|
|
|
bit_buf |= value >> (n - bit_left);
|
|
|
|
if (s->buf_end - s->buf_ptr >= sizeof(BitBuf)) {
|
|
|
|
AV_WBBUF(s->buf_ptr, bit_buf);
|
|
|
|
s->buf_ptr += sizeof(BitBuf);
|
|
|
|
} else {
|
|
|
|
av_log(NULL, AV_LOG_ERROR, "Internal error, put_bits buffer too small\n");
|
|
|
|
av_assert2(0);
|
|
|
|
}
|
|
|
|
bit_left += BUF_BITS - n;
|
|
|
|
bit_buf = value;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
s->bit_buf = bit_buf;
|
|
|
|
s->bit_left = bit_left;
|
|
|
|
}
|
|
|
|
|
avcodec/put_bits: Make bit buffers 64-bit
Change BitBuf into uint64_t on 64-bit x86. This means we need to flush the
buffer less often, which is a significant speed win. All other platforms,
including all 32-bit ones, are unchanged. Output bitstream is the same.
All API constraints are kept in place, e.g., you still cannot put_bits()
more than 31 bits at a time. This is so that codecs cannot accidentally
become 64-bit-only or similar.
Benchmarking on transcoding to various formats shows consistently
positive results:
dnxhd 25.60 fps -> 26.26 fps ( +2.6%)
dvvideo 24.88 fps -> 25.17 fps ( +1.2%)
ffv1 14.32 fps -> 14.58 fps ( +1.8%)
huffyuv 58.75 fps -> 63.27 fps ( +7.7%)
jpegls 6.22 fps -> 6.34 fps ( +1.8%)
magicyuv 57.10 fps -> 63.29 fps (+10.8%)
mjpeg 48.65 fps -> 49.01 fps ( +0.7%)
mpeg1video 76.41 fps -> 77.01 fps ( +0.8%)
mpeg2video 75.99 fps -> 77.43 fps ( +1.9%)
mpeg4 80.66 fps -> 81.37 fps ( +0.9%)
prores 12.35 fps -> 12.88 fps ( +4.3%)
prores_ks 16.20 fps -> 16.80 fps ( +3.7%)
rv20 62.80 fps -> 62.99 fps ( +0.3%)
utvideo 68.41 fps -> 76.32 fps (+11.6%)
Note that this includes video decoding and all other encoding work,
such as DCTs. If you isolate the actual bit-writing routines, it is
likely to be much more.
Benchmark details: Transcoding the first 30 seconds of Big Buck Bunny
in 1080p, Haswell 2.1 GHz, GCC 8.3, generally quantizer locked to
5.0. (Exceptions: DNxHD needs fixed bitrate, and JPEG-LS is so slow
that I only took the first 10 seconds, not 30.) All runs were done
ten times and single-threaded, top and bottom two results discarded to
get rid of outliers, arithmetic mean between the remaining six.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
4 years ago
|
|
|
/**
|
|
|
|
* Write up to 31 bits into a bitstream.
|
|
|
|
* Use put_bits32 to write 32 bits.
|
|
|
|
*/
|
|
|
|
static inline void put_bits(PutBitContext *s, int n, BitBuf value)
|
|
|
|
{
|
|
|
|
av_assert2(n <= 31 && value < (1UL << n));
|
|
|
|
put_bits_no_assert(s, n, value);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void put_bits_le(PutBitContext *s, int n, BitBuf value)
|
|
|
|
{
|
|
|
|
BitBuf bit_buf;
|
|
|
|
int bit_left;
|
|
|
|
|
|
|
|
av_assert2(n <= 31 && value < (1UL << n));
|
|
|
|
|
|
|
|
bit_buf = s->bit_buf;
|
|
|
|
bit_left = s->bit_left;
|
|
|
|
|
|
|
|
bit_buf |= value << (BUF_BITS - bit_left);
|
|
|
|
if (n >= bit_left) {
|
|
|
|
if (s->buf_end - s->buf_ptr >= sizeof(BitBuf)) {
|
|
|
|
AV_WLBUF(s->buf_ptr, bit_buf);
|
|
|
|
s->buf_ptr += sizeof(BitBuf);
|
|
|
|
} else {
|
|
|
|
av_log(NULL, AV_LOG_ERROR, "Internal error, put_bits buffer too small\n");
|
|
|
|
av_assert2(0);
|
|
|
|
}
|
|
|
|
bit_buf = value >> bit_left;
|
|
|
|
bit_left += BUF_BITS;
|
|
|
|
}
|
|
|
|
bit_left -= n;
|
|
|
|
|
|
|
|
s->bit_buf = bit_buf;
|
|
|
|
s->bit_left = bit_left;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void put_sbits(PutBitContext *pb, int n, int32_t value)
|
|
|
|
{
|
|
|
|
av_assert2(n >= 0 && n <= 31);
|
|
|
|
|
|
|
|
put_bits(pb, n, av_mod_uintp2(value, n));
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Write exactly 32 bits into a bitstream.
|
|
|
|
*/
|
|
|
|
static void av_unused put_bits32(PutBitContext *s, uint32_t value)
|
|
|
|
{
|
|
|
|
BitBuf bit_buf;
|
|
|
|
int bit_left;
|
|
|
|
|
avcodec/put_bits: Make bit buffers 64-bit
Change BitBuf into uint64_t on 64-bit x86. This means we need to flush the
buffer less often, which is a significant speed win. All other platforms,
including all 32-bit ones, are unchanged. Output bitstream is the same.
All API constraints are kept in place, e.g., you still cannot put_bits()
more than 31 bits at a time. This is so that codecs cannot accidentally
become 64-bit-only or similar.
Benchmarking on transcoding to various formats shows consistently
positive results:
dnxhd 25.60 fps -> 26.26 fps ( +2.6%)
dvvideo 24.88 fps -> 25.17 fps ( +1.2%)
ffv1 14.32 fps -> 14.58 fps ( +1.8%)
huffyuv 58.75 fps -> 63.27 fps ( +7.7%)
jpegls 6.22 fps -> 6.34 fps ( +1.8%)
magicyuv 57.10 fps -> 63.29 fps (+10.8%)
mjpeg 48.65 fps -> 49.01 fps ( +0.7%)
mpeg1video 76.41 fps -> 77.01 fps ( +0.8%)
mpeg2video 75.99 fps -> 77.43 fps ( +1.9%)
mpeg4 80.66 fps -> 81.37 fps ( +0.9%)
prores 12.35 fps -> 12.88 fps ( +4.3%)
prores_ks 16.20 fps -> 16.80 fps ( +3.7%)
rv20 62.80 fps -> 62.99 fps ( +0.3%)
utvideo 68.41 fps -> 76.32 fps (+11.6%)
Note that this includes video decoding and all other encoding work,
such as DCTs. If you isolate the actual bit-writing routines, it is
likely to be much more.
Benchmark details: Transcoding the first 30 seconds of Big Buck Bunny
in 1080p, Haswell 2.1 GHz, GCC 8.3, generally quantizer locked to
5.0. (Exceptions: DNxHD needs fixed bitrate, and JPEG-LS is so slow
that I only took the first 10 seconds, not 30.) All runs were done
ten times and single-threaded, top and bottom two results discarded to
get rid of outliers, arithmetic mean between the remaining six.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
4 years ago
|
|
|
if (BUF_BITS > 32) {
|
|
|
|
put_bits_no_assert(s, 32, value);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
bit_buf = s->bit_buf;
|
|
|
|
bit_left = s->bit_left;
|
|
|
|
|
|
|
|
#ifdef BITSTREAM_WRITER_LE
|
|
|
|
bit_buf |= (BitBuf)value << (BUF_BITS - bit_left);
|
|
|
|
if (s->buf_end - s->buf_ptr >= sizeof(BitBuf)) {
|
|
|
|
AV_WLBUF(s->buf_ptr, bit_buf);
|
|
|
|
s->buf_ptr += sizeof(BitBuf);
|
|
|
|
} else {
|
|
|
|
av_log(NULL, AV_LOG_ERROR, "Internal error, put_bits buffer too small\n");
|
|
|
|
av_assert2(0);
|
|
|
|
}
|
|
|
|
bit_buf = (uint64_t)value >> bit_left;
|
|
|
|
#else
|
|
|
|
bit_buf = (uint64_t)bit_buf << bit_left;
|
|
|
|
bit_buf |= (BitBuf)value >> (BUF_BITS - bit_left);
|
|
|
|
if (s->buf_end - s->buf_ptr >= sizeof(BitBuf)) {
|
|
|
|
AV_WBBUF(s->buf_ptr, bit_buf);
|
|
|
|
s->buf_ptr += sizeof(BitBuf);
|
|
|
|
} else {
|
|
|
|
av_log(NULL, AV_LOG_ERROR, "Internal error, put_bits buffer too small\n");
|
|
|
|
av_assert2(0);
|
|
|
|
}
|
|
|
|
bit_buf = value;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
s->bit_buf = bit_buf;
|
|
|
|
s->bit_left = bit_left;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Write up to 64 bits into a bitstream.
|
|
|
|
*/
|
|
|
|
static inline void put_bits64(PutBitContext *s, int n, uint64_t value)
|
|
|
|
{
|
|
|
|
av_assert2((n == 64) || (n < 64 && value < (UINT64_C(1) << n)));
|
|
|
|
|
|
|
|
if (n < 32)
|
|
|
|
put_bits(s, n, value);
|
|
|
|
else if (n == 32)
|
|
|
|
put_bits32(s, value);
|
|
|
|
else if (n < 64) {
|
|
|
|
uint32_t lo = value & 0xffffffff;
|
|
|
|
uint32_t hi = value >> 32;
|
|
|
|
#ifdef BITSTREAM_WRITER_LE
|
|
|
|
put_bits32(s, lo);
|
|
|
|
put_bits(s, n - 32, hi);
|
|
|
|
#else
|
|
|
|
put_bits(s, n - 32, hi);
|
|
|
|
put_bits32(s, lo);
|
|
|
|
#endif
|
|
|
|
} else {
|
|
|
|
uint32_t lo = value & 0xffffffff;
|
|
|
|
uint32_t hi = value >> 32;
|
|
|
|
#ifdef BITSTREAM_WRITER_LE
|
|
|
|
put_bits32(s, lo);
|
|
|
|
put_bits32(s, hi);
|
|
|
|
#else
|
|
|
|
put_bits32(s, hi);
|
|
|
|
put_bits32(s, lo);
|
|
|
|
#endif
|
|
|
|
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Return the pointer to the byte where the bitstream writer will put
|
|
|
|
* the next bit.
|
|
|
|
*/
|
|
|
|
static inline uint8_t *put_bits_ptr(PutBitContext *s)
|
|
|
|
{
|
|
|
|
return s->buf_ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Skip the given number of bytes.
|
|
|
|
* PutBitContext must be flushed & aligned to a byte boundary before calling this.
|
|
|
|
*/
|
|
|
|
static inline void skip_put_bytes(PutBitContext *s, int n)
|
|
|
|
{
|
|
|
|
av_assert2((put_bits_count(s) & 7) == 0);
|
|
|
|
av_assert2(s->bit_left == BUF_BITS);
|
|
|
|
av_assert0(n <= s->buf_end - s->buf_ptr);
|
|
|
|
s->buf_ptr += n;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Skip the given number of bits.
|
|
|
|
* Must only be used if the actual values in the bitstream do not matter.
|
|
|
|
* If n is < 0 the behavior is undefined.
|
|
|
|
*/
|
|
|
|
static inline void skip_put_bits(PutBitContext *s, int n)
|
|
|
|
{
|
|
|
|
unsigned bits = BUF_BITS - s->bit_left + n;
|
|
|
|
s->buf_ptr += sizeof(BitBuf) * (bits / BUF_BITS);
|
|
|
|
s->bit_left = BUF_BITS - (bits & (BUF_BITS - 1));
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Change the end of the buffer.
|
|
|
|
*
|
|
|
|
* @param size the new size in bytes of the buffer where to put bits
|
|
|
|
*/
|
|
|
|
static inline void set_put_bits_buffer_size(PutBitContext *s, int size)
|
|
|
|
{
|
|
|
|
av_assert0(size <= INT_MAX/8 - BUF_BITS);
|
|
|
|
s->buf_end = s->buf + size;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Pad the bitstream with zeros up to the next byte boundary.
|
|
|
|
*/
|
|
|
|
static inline void align_put_bits(PutBitContext *s)
|
|
|
|
{
|
|
|
|
put_bits(s, s->bit_left & 7, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
#undef AV_WBBUF
|
|
|
|
#undef AV_WLBUF
|
|
|
|
|
|
|
|
#endif /* AVCODEC_PUT_BITS_H */
|