2011-06-10 21:57:19 +03:00
|
|
|
/*
|
|
|
|
* AC-3 encoder float/fixed template
|
|
|
|
* Copyright (c) 2000 Fabrice Bellard
|
|
|
|
* Copyright (c) 2006-2011 Justin Ruggles <justin.ruggles@gmail.com>
|
|
|
|
* Copyright (c) 2006-2010 Prakash Punnoor <prakash@punnoor.de>
|
|
|
|
*
|
2012-10-07 12:23:29 +03:00
|
|
|
* This file is part of FFmpeg.
|
2011-06-10 21:57:19 +03:00
|
|
|
*
|
2012-10-07 12:23:29 +03:00
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2011-06-10 21:57:19 +03:00
|
|
|
* 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.
|
|
|
|
*
|
2012-10-07 12:23:29 +03:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2011-06-10 21:57:19 +03:00
|
|
|
* 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
|
2012-10-07 12:23:29 +03:00
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
2011-06-10 21:57:19 +03:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @file
|
|
|
|
* AC-3 encoder float/fixed template
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <stdint.h>
|
|
|
|
|
2014-01-09 12:27:20 +03:00
|
|
|
#include "libavutil/attributes.h"
|
2013-02-08 02:31:13 +03:00
|
|
|
#include "libavutil/internal.h"
|
2020-06-10 15:38:08 +02:00
|
|
|
#include "libavutil/mem_internal.h"
|
2014-01-16 19:30:19 +03:00
|
|
|
|
|
|
|
#include "audiodsp.h"
|
2014-01-09 12:27:20 +03:00
|
|
|
#include "internal.h"
|
|
|
|
#include "ac3enc.h"
|
|
|
|
#include "eac3enc.h"
|
2011-06-10 21:57:19 +03:00
|
|
|
|
2011-06-27 06:58:19 +03:00
|
|
|
|
2020-12-03 04:36:32 +02:00
|
|
|
static int allocate_sample_buffers(AC3EncodeContext *s)
|
2011-06-14 00:44:50 +03:00
|
|
|
{
|
|
|
|
int ch;
|
|
|
|
|
2020-05-29 16:05:46 +02:00
|
|
|
if (!FF_ALLOC_TYPED_ARRAY(s->windowed_samples, AC3_WINDOW_SIZE) ||
|
2020-09-14 02:59:51 +02:00
|
|
|
!FF_ALLOCZ_TYPED_ARRAY(s->planar_samples, s->channels))
|
2020-05-29 16:05:46 +02:00
|
|
|
return AVERROR(ENOMEM);
|
|
|
|
|
2011-06-14 00:44:50 +03:00
|
|
|
for (ch = 0; ch < s->channels; ch++) {
|
2020-05-29 16:05:46 +02:00
|
|
|
if (!(s->planar_samples[ch] = av_mallocz((AC3_FRAME_SIZE + AC3_BLOCK_SIZE) *
|
|
|
|
sizeof(**s->planar_samples))))
|
|
|
|
return AVERROR(ENOMEM);
|
2011-06-14 00:44:50 +03:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2011-11-02 17:38:10 +03:00
|
|
|
/*
|
2012-08-25 20:11:10 +03:00
|
|
|
* Copy input samples.
|
2011-11-29 17:51:36 +03:00
|
|
|
* Channels are reordered from FFmpeg's default order to AC-3 order.
|
2011-06-10 21:57:19 +03:00
|
|
|
*/
|
2012-08-25 20:11:10 +03:00
|
|
|
static void copy_input_samples(AC3EncodeContext *s, SampleType **samples)
|
2011-06-10 21:57:19 +03:00
|
|
|
{
|
2012-08-25 20:11:10 +03:00
|
|
|
int ch;
|
2011-06-10 21:57:19 +03:00
|
|
|
|
2012-08-25 20:11:10 +03:00
|
|
|
/* copy and remap input samples */
|
2011-06-10 21:57:19 +03:00
|
|
|
for (ch = 0; ch < s->channels; ch++) {
|
|
|
|
/* copy last 256 samples of previous frame to the start of the current frame */
|
2011-07-14 20:02:45 +03:00
|
|
|
memcpy(&s->planar_samples[ch][0], &s->planar_samples[ch][AC3_BLOCK_SIZE * s->num_blocks],
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3_BLOCK_SIZE * sizeof(s->planar_samples[0][0]));
|
|
|
|
|
2012-08-25 20:11:10 +03:00
|
|
|
/* copy new samples for current frame */
|
|
|
|
memcpy(&s->planar_samples[ch][AC3_BLOCK_SIZE],
|
|
|
|
samples[s->channel_map[ch]],
|
|
|
|
AC3_BLOCK_SIZE * s->num_blocks * sizeof(s->planar_samples[0][0]));
|
2011-06-10 21:57:19 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2011-11-02 17:38:10 +03:00
|
|
|
/*
|
2011-06-10 21:57:19 +03:00
|
|
|
* Apply the MDCT to input samples to generate frequency coefficients.
|
|
|
|
* This applies the KBD window and normalizes the input to reduce precision
|
|
|
|
* loss due to fixed-point calculations.
|
|
|
|
*/
|
2011-06-27 06:58:19 +03:00
|
|
|
static void apply_mdct(AC3EncodeContext *s)
|
2011-06-10 21:57:19 +03:00
|
|
|
{
|
|
|
|
int blk, ch;
|
|
|
|
|
|
|
|
for (ch = 0; ch < s->channels; ch++) {
|
2011-07-14 20:02:45 +03:00
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3Block *block = &s->blocks[blk];
|
|
|
|
const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
|
|
|
|
|
2014-11-29 19:54:27 +02:00
|
|
|
s->fdsp->vector_fmul(s->windowed_samples, input_samples,
|
2021-01-09 18:27:16 +02:00
|
|
|
s->mdct_window, AC3_BLOCK_SIZE);
|
|
|
|
s->fdsp->vector_fmul_reverse(s->windowed_samples + AC3_BLOCK_SIZE,
|
|
|
|
&input_samples[AC3_BLOCK_SIZE],
|
|
|
|
s->mdct_window, AC3_BLOCK_SIZE);
|
2011-06-10 21:57:19 +03:00
|
|
|
|
ac3enc_fixed: convert to 32-bit sample format
The AC3 encoder used to be a separate library called "Aften", which
got merged into libavcodec (literally, SVN commits and all).
The merge preserved as much features from the library as possible.
The code had two versions - a fixed point version and a floating
point version. FFmpeg had floating point DSP code used by other
codecs, the AC3 decoder including, so the floating-point DSP was
simply replaced with FFmpeg's own functions.
However, FFmpeg had no fixed-point audio code at that point. So
the encoder brought along its own fixed-point DSP functions,
including a fixed-point MDCT.
The fixed-point MDCT itself is trivially just a float MDCT with a
different type and each multiply being a fixed-point multiply.
So over time, it got refactored, and the FFT used for all other codecs
was templated.
Due to design decisions at the time, the fixed-point version of the
encoder operates at 16-bits of precision. Although convenient, this,
even at the time, was inadequate and inefficient. The encoder is noisy,
does not produce output comparable to the float encoder, and even
rings at higher frequencies due to the badly approximated winow function.
Enter MIPS (owned by Imagination Technologies at the time). They wanted
quick fixed-point decoding on their FPUless cores. So they contributed
patches to template the AC3 decoder so it had both a fixed-point
and a floating-point version. They also did the same for the AAC decoder.
They however, used 32-bit samples. Not 16-bits. And we did not have
32-bit fixed-point DSP functions, including an MDCT. But instead of
templating our MDCT to output 3 versions (float, 32-bit fixed and 16-bit fixed),
they simply copy-pasted their own MDCT into ours, and completely
ifdeffed our own MDCT code out if a 32-bit fixed point MDCT was selected.
This is also the status quo nowadays - 2 separate MDCTs, one which
produces floating point and 16-bit fixed point versions, and one
sort-of integrated which produces 32-bit MDCT.
MIPS weren't all that interested in encoding, so they left the encoder
as-is, and they didn't care much about the ifdeffery, mess or quality - it's
not their problem.
So the MDCT/FFT code has always been a thorn in anyone looking to clean up
code's eye.
Backstory over. Internally AC3 operates on 25-bit fixed-point coefficients.
So for the floating point version, the encoder simply runs the float MDCT,
and converts the resulting coefficients to 25-bit fixed-point, as AC3 is inherently
a fixed-point codec. For the fixed-point version, the input is 16-bit samples,
so to maximize precision the frame samples are analyzed and the highest set
bit is detected via ac3_max_msb_abs_int16(), and the coefficients are then
scaled up via ac3_lshift_int16(), so the input for the FFT is always at least 14 bits,
computed in normalize_samples(). After FFT, the coefficients are scaled up to 25 bits.
This patch simply changes the encoder to accept 32-bit samples, reusing
the already well-optimized 32-bit MDCT code, allowing us to clean up and drop
a large part of a very messy code of ours, as well as prepare for the future lavu/tx
conversion. The coefficients are simply scaled down to 25 bits during windowing,
skipping 2 separate scalings, as the hacks to extend precision are simply no longer
necessary. There's no point in running the MDCT always at 32 bits when you're
going to drop 6 bits off anyway, the headroom is plenty, and the MDCT rounds
properly.
This also makes the encoder even slightly more accurate over the float version,
as there's no coefficient conversion step necessary.
SIZE SAVINGS:
ARM32:
HARDCODED TABLES:
BASE - 10709590
DROP DSP - 10702872 - diff: -6.56KiB
DROP MDCT - 10667932 - diff: -34.12KiB - both: -40.68KiB
DROP FFT - 10336652 - diff: -323.52KiB - all: -364.20KiB
SOFTCODED TABLES:
BASE - 9685096
DROP DSP - 9678378 - diff: -6.56KiB
DROP MDCT - 9643466 - diff: -34.09KiB - both: -40.65KiB
DROP FFT - 9573918 - diff: -67.92KiB - all: -108.57KiB
ARM64:
HARDCODED TABLES:
BASE - 14641112
DROP DSP - 14633806 - diff: -7.13KiB
DROP MDCT - 14604812 - diff: -28.31KiB - both: -35.45KiB
DROP FFT - 14286826 - diff: -310.53KiB - all: -345.98KiB
SOFTCODED TABLES:
BASE - 13636238
DROP DSP - 13628932 - diff: -7.13KiB
DROP MDCT - 13599866 - diff: -28.38KiB - both: -35.52KiB
DROP FFT - 13542080 - diff: -56.43KiB - all: -91.95KiB
x86:
HARDCODED TABLES:
BASE - 12367336
DROP DSP - 12354698 - diff: -12.34KiB
DROP MDCT - 12331024 - diff: -23.12KiB - both: -35.46KiB
DROP FFT - 12029788 - diff: -294.18KiB - all: -329.64KiB
SOFTCODED TABLES:
BASE - 11358094
DROP DSP - 11345456 - diff: -12.34KiB
DROP MDCT - 11321742 - diff: -23.16KiB - both: -35.50KiB
DROP FFT - 11276946 - diff: -43.75KiB - all: -79.25KiB
PERFORMANCE (10min random s32le):
ARM32 - before - 39.9x - 0m15.046s
ARM32 - after - 28.2x - 0m21.525s
Speed: -30%
ARM64 - before - 36.1x - 0m16.637s
ARM64 - after - 36.0x - 0m16.727s
Speed: -0.5%
x86 - before - 184x - 0m3.277s
x86 - after - 190x - 0m3.187s
Speed: +3%
2021-01-09 02:51:52 +02:00
|
|
|
s->mdct.mdct_calc(&s->mdct, block->mdct_coef[ch+1],
|
|
|
|
s->windowed_samples);
|
2011-06-10 21:57:19 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2011-11-02 17:38:10 +03:00
|
|
|
/*
|
2011-06-10 21:57:19 +03:00
|
|
|
* Calculate coupling channel and coupling coordinates.
|
|
|
|
*/
|
2011-06-27 06:58:19 +03:00
|
|
|
static void apply_channel_coupling(AC3EncodeContext *s)
|
2011-06-10 21:57:19 +03:00
|
|
|
{
|
2011-08-08 00:47:42 +03:00
|
|
|
LOCAL_ALIGNED_16(CoefType, cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
|
2020-12-03 05:13:28 +02:00
|
|
|
#if AC3ENC_FLOAT
|
2011-06-10 21:57:19 +03:00
|
|
|
LOCAL_ALIGNED_16(int32_t, fixed_cpl_coords, [AC3_MAX_BLOCKS], [AC3_MAX_CHANNELS][16]);
|
2011-08-08 00:47:42 +03:00
|
|
|
#else
|
|
|
|
int32_t (*fixed_cpl_coords)[AC3_MAX_CHANNELS][16] = cpl_coords;
|
|
|
|
#endif
|
2013-02-26 04:32:44 +03:00
|
|
|
int av_uninit(blk), ch, bnd, i, j;
|
2011-06-10 21:57:19 +03:00
|
|
|
CoefSumType energy[AC3_MAX_BLOCKS][AC3_MAX_CHANNELS][16] = {{{0}}};
|
2011-06-14 20:45:38 +03:00
|
|
|
int cpl_start, num_cpl_coefs;
|
2011-06-10 21:57:19 +03:00
|
|
|
|
|
|
|
memset(cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
|
2020-12-03 05:13:28 +02:00
|
|
|
#if AC3ENC_FLOAT
|
2011-08-08 00:47:42 +03:00
|
|
|
memset(fixed_cpl_coords, 0, AC3_MAX_BLOCKS * sizeof(*cpl_coords));
|
|
|
|
#endif
|
2011-06-10 21:57:19 +03:00
|
|
|
|
2011-06-14 20:45:38 +03:00
|
|
|
/* align start to 16-byte boundary. align length to multiple of 32.
|
|
|
|
note: coupling start bin % 4 will always be 1 */
|
|
|
|
cpl_start = s->start_freq[CPL_CH] - 1;
|
|
|
|
num_cpl_coefs = FFALIGN(s->num_cpl_subbands * 12 + 1, 32);
|
|
|
|
cpl_start = FFMIN(256, cpl_start + num_cpl_coefs) - num_cpl_coefs;
|
|
|
|
|
2011-06-10 21:57:19 +03:00
|
|
|
/* calculate coupling channel from fbw channels */
|
2011-07-14 20:02:45 +03:00
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3Block *block = &s->blocks[blk];
|
2011-06-14 20:45:38 +03:00
|
|
|
CoefType *cpl_coef = &block->mdct_coef[CPL_CH][cpl_start];
|
2011-06-10 21:57:19 +03:00
|
|
|
if (!block->cpl_in_use)
|
|
|
|
continue;
|
2011-06-14 20:45:38 +03:00
|
|
|
memset(cpl_coef, 0, num_cpl_coefs * sizeof(*cpl_coef));
|
2011-06-10 21:57:19 +03:00
|
|
|
for (ch = 1; ch <= s->fbw_channels; ch++) {
|
2011-06-14 20:45:38 +03:00
|
|
|
CoefType *ch_coef = &block->mdct_coef[ch][cpl_start];
|
2011-06-10 21:57:19 +03:00
|
|
|
if (!block->channel_in_cpl[ch])
|
|
|
|
continue;
|
|
|
|
for (i = 0; i < num_cpl_coefs; i++)
|
|
|
|
cpl_coef[i] += ch_coef[i];
|
|
|
|
}
|
|
|
|
|
2011-06-27 21:29:33 +03:00
|
|
|
/* coefficients must be clipped in order to be encoded */
|
2014-01-16 19:30:19 +03:00
|
|
|
clip_coefficients(&s->adsp, cpl_coef, num_cpl_coefs);
|
2011-06-10 21:57:19 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
/* calculate energy in each band in coupling channel and each fbw channel */
|
|
|
|
/* TODO: possibly use SIMD to speed up energy calculation */
|
|
|
|
bnd = 0;
|
|
|
|
i = s->start_freq[CPL_CH];
|
|
|
|
while (i < s->cpl_end_freq) {
|
|
|
|
int band_size = s->cpl_band_sizes[bnd];
|
|
|
|
for (ch = CPL_CH; ch <= s->fbw_channels; ch++) {
|
2011-07-14 20:02:45 +03:00
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3Block *block = &s->blocks[blk];
|
|
|
|
if (!block->cpl_in_use || (ch > CPL_CH && !block->channel_in_cpl[ch]))
|
|
|
|
continue;
|
|
|
|
for (j = 0; j < band_size; j++) {
|
|
|
|
CoefType v = block->mdct_coef[ch][i+j];
|
|
|
|
MAC_COEF(energy[blk][ch][bnd], v, v);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
i += band_size;
|
|
|
|
bnd++;
|
|
|
|
}
|
|
|
|
|
2011-08-06 00:34:53 +03:00
|
|
|
/* calculate coupling coordinates for all blocks for all channels */
|
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
|
|
|
AC3Block *block = &s->blocks[blk];
|
|
|
|
if (!block->cpl_in_use)
|
|
|
|
continue;
|
|
|
|
for (ch = 1; ch <= s->fbw_channels; ch++) {
|
|
|
|
if (!block->channel_in_cpl[ch])
|
|
|
|
continue;
|
|
|
|
for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
|
|
|
|
cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy[blk][ch][bnd],
|
|
|
|
energy[blk][CPL_CH][bnd]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-06-10 21:57:19 +03:00
|
|
|
/* determine which blocks to send new coupling coordinates for */
|
2011-07-14 20:02:45 +03:00
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3Block *block = &s->blocks[blk];
|
|
|
|
AC3Block *block0 = blk ? &s->blocks[blk-1] : NULL;
|
|
|
|
|
2011-08-05 23:28:39 +03:00
|
|
|
memset(block->new_cpl_coords, 0, sizeof(block->new_cpl_coords));
|
|
|
|
|
2011-06-10 21:57:19 +03:00
|
|
|
if (block->cpl_in_use) {
|
|
|
|
/* send new coordinates if this is the first block, if previous
|
|
|
|
* block did not use coupling but this block does, the channels
|
|
|
|
* using coupling has changed from the previous block, or the
|
|
|
|
* coordinate difference from the last block for any channel is
|
|
|
|
* greater than a threshold value. */
|
2011-08-05 23:28:39 +03:00
|
|
|
if (blk == 0 || !block0->cpl_in_use) {
|
|
|
|
for (ch = 1; ch <= s->fbw_channels; ch++)
|
|
|
|
block->new_cpl_coords[ch] = 1;
|
2011-06-10 21:57:19 +03:00
|
|
|
} else {
|
|
|
|
for (ch = 1; ch <= s->fbw_channels; ch++) {
|
2011-08-06 00:34:53 +03:00
|
|
|
if (!block->channel_in_cpl[ch])
|
|
|
|
continue;
|
|
|
|
if (!block0->channel_in_cpl[ch]) {
|
2011-08-05 23:28:39 +03:00
|
|
|
block->new_cpl_coords[ch] = 1;
|
2011-08-06 00:34:53 +03:00
|
|
|
} else {
|
|
|
|
CoefSumType coord_diff = 0;
|
|
|
|
for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
|
2011-08-08 00:47:42 +03:00
|
|
|
coord_diff += FFABS(cpl_coords[blk-1][ch][bnd] -
|
|
|
|
cpl_coords[blk ][ch][bnd]);
|
2011-08-06 00:34:53 +03:00
|
|
|
}
|
|
|
|
coord_diff /= s->num_cpl_bands;
|
2011-08-08 00:47:42 +03:00
|
|
|
if (coord_diff > NEW_CPL_COORD_THRESHOLD)
|
2011-08-06 00:34:53 +03:00
|
|
|
block->new_cpl_coords[ch] = 1;
|
2011-06-10 21:57:19 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* calculate final coupling coordinates, taking into account reusing of
|
|
|
|
coordinates in successive blocks */
|
|
|
|
for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
|
|
|
|
blk = 0;
|
2011-07-14 20:02:45 +03:00
|
|
|
while (blk < s->num_blocks) {
|
2011-08-05 23:28:39 +03:00
|
|
|
int av_uninit(blk1);
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3Block *block = &s->blocks[blk];
|
|
|
|
|
|
|
|
if (!block->cpl_in_use) {
|
|
|
|
blk++;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (ch = 1; ch <= s->fbw_channels; ch++) {
|
2011-08-05 23:28:39 +03:00
|
|
|
CoefSumType energy_ch, energy_cpl;
|
2011-06-10 21:57:19 +03:00
|
|
|
if (!block->channel_in_cpl[ch])
|
|
|
|
continue;
|
2011-08-05 23:28:39 +03:00
|
|
|
energy_cpl = energy[blk][CPL_CH][bnd];
|
2011-06-10 21:57:19 +03:00
|
|
|
energy_ch = energy[blk][ch][bnd];
|
|
|
|
blk1 = blk+1;
|
2014-09-27 21:34:44 +03:00
|
|
|
while (blk1 < s->num_blocks && !s->blocks[blk1].new_cpl_coords[ch]) {
|
2011-08-05 23:28:39 +03:00
|
|
|
if (s->blocks[blk1].cpl_in_use) {
|
|
|
|
energy_cpl += energy[blk1][CPL_CH][bnd];
|
2011-06-10 21:57:19 +03:00
|
|
|
energy_ch += energy[blk1][ch][bnd];
|
2011-08-05 23:28:39 +03:00
|
|
|
}
|
2011-06-10 21:57:19 +03:00
|
|
|
blk1++;
|
|
|
|
}
|
|
|
|
cpl_coords[blk][ch][bnd] = calc_cpl_coord(energy_ch, energy_cpl);
|
|
|
|
}
|
|
|
|
blk = blk1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* calculate exponents/mantissas for coupling coordinates */
|
2011-07-14 20:02:45 +03:00
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
2011-06-10 21:57:19 +03:00
|
|
|
AC3Block *block = &s->blocks[blk];
|
2011-08-05 23:28:39 +03:00
|
|
|
if (!block->cpl_in_use)
|
2011-06-10 21:57:19 +03:00
|
|
|
continue;
|
|
|
|
|
2020-12-03 05:13:28 +02:00
|
|
|
#if AC3ENC_FLOAT
|
2011-06-10 21:57:19 +03:00
|
|
|
s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
|
|
|
|
cpl_coords[blk][1],
|
|
|
|
s->fbw_channels * 16);
|
2011-08-08 00:47:42 +03:00
|
|
|
#endif
|
2011-06-10 21:57:19 +03:00
|
|
|
s->ac3dsp.extract_exponents(block->cpl_coord_exp[1],
|
|
|
|
fixed_cpl_coords[blk][1],
|
|
|
|
s->fbw_channels * 16);
|
|
|
|
|
|
|
|
for (ch = 1; ch <= s->fbw_channels; ch++) {
|
|
|
|
int bnd, min_exp, max_exp, master_exp;
|
|
|
|
|
2011-08-05 23:28:39 +03:00
|
|
|
if (!block->new_cpl_coords[ch])
|
|
|
|
continue;
|
|
|
|
|
2011-06-10 21:57:19 +03:00
|
|
|
/* determine master exponent */
|
|
|
|
min_exp = max_exp = block->cpl_coord_exp[ch][0];
|
|
|
|
for (bnd = 1; bnd < s->num_cpl_bands; bnd++) {
|
|
|
|
int exp = block->cpl_coord_exp[ch][bnd];
|
|
|
|
min_exp = FFMIN(exp, min_exp);
|
|
|
|
max_exp = FFMAX(exp, max_exp);
|
|
|
|
}
|
|
|
|
master_exp = ((max_exp - 15) + 2) / 3;
|
|
|
|
master_exp = FFMAX(master_exp, 0);
|
|
|
|
while (min_exp < master_exp * 3)
|
|
|
|
master_exp--;
|
|
|
|
for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
|
|
|
|
block->cpl_coord_exp[ch][bnd] = av_clip(block->cpl_coord_exp[ch][bnd] -
|
|
|
|
master_exp * 3, 0, 15);
|
|
|
|
}
|
|
|
|
block->cpl_master_exp[ch] = master_exp;
|
|
|
|
|
|
|
|
/* quantize mantissas */
|
|
|
|
for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
|
|
|
|
int cpl_exp = block->cpl_coord_exp[ch][bnd];
|
|
|
|
int cpl_mant = (fixed_cpl_coords[blk][ch][bnd] << (5 + cpl_exp + master_exp * 3)) >> 24;
|
|
|
|
if (cpl_exp == 15)
|
|
|
|
cpl_mant >>= 1;
|
|
|
|
else
|
|
|
|
cpl_mant -= 16;
|
|
|
|
|
|
|
|
block->cpl_coord_mant[ch][bnd] = cpl_mant;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-12-03 05:44:21 +02:00
|
|
|
if (AC3ENC_FLOAT && CONFIG_EAC3_ENCODER && s->eac3)
|
2011-06-10 21:57:19 +03:00
|
|
|
ff_eac3_set_cpl_states(s);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2011-11-02 17:38:10 +03:00
|
|
|
/*
|
2011-06-10 21:57:19 +03:00
|
|
|
* Determine rematrixing flags for each block and band.
|
|
|
|
*/
|
2011-06-27 06:58:19 +03:00
|
|
|
static void compute_rematrixing_strategy(AC3EncodeContext *s)
|
2011-06-10 21:57:19 +03:00
|
|
|
{
|
|
|
|
int nb_coefs;
|
2011-04-03 21:38:41 +03:00
|
|
|
int blk, bnd;
|
2013-01-29 18:15:02 +03:00
|
|
|
AC3Block *block, *block0 = NULL;
|
2011-06-10 21:57:19 +03:00
|
|
|
|
|
|
|
if (s->channel_mode != AC3_CHMODE_STEREO)
|
|
|
|
return;
|
|
|
|
|
2011-07-14 20:02:45 +03:00
|
|
|
for (blk = 0; blk < s->num_blocks; blk++) {
|
2011-06-10 21:57:19 +03:00
|
|
|
block = &s->blocks[blk];
|
|
|
|
block->new_rematrixing_strategy = !blk;
|
|
|
|
|
|
|
|
block->num_rematrixing_bands = 4;
|
|
|
|
if (block->cpl_in_use) {
|
|
|
|
block->num_rematrixing_bands -= (s->start_freq[CPL_CH] <= 61);
|
|
|
|
block->num_rematrixing_bands -= (s->start_freq[CPL_CH] == 37);
|
|
|
|
if (blk && block->num_rematrixing_bands != block0->num_rematrixing_bands)
|
|
|
|
block->new_rematrixing_strategy = 1;
|
|
|
|
}
|
|
|
|
nb_coefs = FFMIN(block->end_freq[1], block->end_freq[2]);
|
|
|
|
|
2011-08-26 20:02:29 +03:00
|
|
|
if (!s->rematrixing_enabled) {
|
|
|
|
block0 = block;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2011-06-10 21:57:19 +03:00
|
|
|
for (bnd = 0; bnd < block->num_rematrixing_bands; bnd++) {
|
2013-08-12 20:40:29 +03:00
|
|
|
/* calculate sum of squared coeffs for one band in one block */
|
2011-06-10 21:57:19 +03:00
|
|
|
int start = ff_ac3_rematrix_band_tab[bnd];
|
|
|
|
int end = FFMIN(nb_coefs, ff_ac3_rematrix_band_tab[bnd+1]);
|
2011-04-03 21:38:41 +03:00
|
|
|
CoefSumType sum[4];
|
|
|
|
sum_square_butterfly(s, sum, block->mdct_coef[1] + start,
|
|
|
|
block->mdct_coef[2] + start, end - start);
|
2011-06-10 21:57:19 +03:00
|
|
|
|
|
|
|
/* compare sums to determine if rematrixing will be used for this band */
|
|
|
|
if (FFMIN(sum[2], sum[3]) < FFMIN(sum[0], sum[1]))
|
|
|
|
block->rematrixing_flags[bnd] = 1;
|
|
|
|
else
|
|
|
|
block->rematrixing_flags[bnd] = 0;
|
|
|
|
|
|
|
|
/* determine if new rematrixing flags will be sent */
|
|
|
|
if (blk &&
|
|
|
|
block->rematrixing_flags[bnd] != block0->rematrixing_flags[bnd]) {
|
|
|
|
block->new_rematrixing_strategy = 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
block0 = block;
|
|
|
|
}
|
|
|
|
}
|
2011-06-27 06:58:19 +03:00
|
|
|
|
|
|
|
|
2012-01-30 03:01:10 +03:00
|
|
|
int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
|
|
|
|
const AVFrame *frame, int *got_packet_ptr)
|
2011-06-27 06:58:19 +03:00
|
|
|
{
|
|
|
|
AC3EncodeContext *s = avctx->priv_data;
|
|
|
|
int ret;
|
|
|
|
|
2011-05-30 02:16:46 +03:00
|
|
|
if (s->options.allow_per_frame_metadata) {
|
2011-07-13 20:53:58 +03:00
|
|
|
ret = ff_ac3_validate_metadata(s);
|
2011-06-27 06:58:19 +03:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2020-12-03 05:44:21 +02:00
|
|
|
if (s->bit_alloc.sr_code == 1 || (AC3ENC_FLOAT && s->eac3))
|
2011-06-27 06:58:19 +03:00
|
|
|
ff_ac3_adjust_frame_size(s);
|
|
|
|
|
2012-08-25 20:11:10 +03:00
|
|
|
copy_input_samples(s, (SampleType **)frame->extended_data);
|
2011-06-27 06:58:19 +03:00
|
|
|
|
|
|
|
apply_mdct(s);
|
|
|
|
|
|
|
|
s->cpl_on = s->cpl_enabled;
|
|
|
|
ff_ac3_compute_coupling_strategy(s);
|
|
|
|
|
|
|
|
if (s->cpl_on)
|
|
|
|
apply_channel_coupling(s);
|
|
|
|
|
|
|
|
compute_rematrixing_strategy(s);
|
|
|
|
|
ac3enc_fixed: convert to 32-bit sample format
The AC3 encoder used to be a separate library called "Aften", which
got merged into libavcodec (literally, SVN commits and all).
The merge preserved as much features from the library as possible.
The code had two versions - a fixed point version and a floating
point version. FFmpeg had floating point DSP code used by other
codecs, the AC3 decoder including, so the floating-point DSP was
simply replaced with FFmpeg's own functions.
However, FFmpeg had no fixed-point audio code at that point. So
the encoder brought along its own fixed-point DSP functions,
including a fixed-point MDCT.
The fixed-point MDCT itself is trivially just a float MDCT with a
different type and each multiply being a fixed-point multiply.
So over time, it got refactored, and the FFT used for all other codecs
was templated.
Due to design decisions at the time, the fixed-point version of the
encoder operates at 16-bits of precision. Although convenient, this,
even at the time, was inadequate and inefficient. The encoder is noisy,
does not produce output comparable to the float encoder, and even
rings at higher frequencies due to the badly approximated winow function.
Enter MIPS (owned by Imagination Technologies at the time). They wanted
quick fixed-point decoding on their FPUless cores. So they contributed
patches to template the AC3 decoder so it had both a fixed-point
and a floating-point version. They also did the same for the AAC decoder.
They however, used 32-bit samples. Not 16-bits. And we did not have
32-bit fixed-point DSP functions, including an MDCT. But instead of
templating our MDCT to output 3 versions (float, 32-bit fixed and 16-bit fixed),
they simply copy-pasted their own MDCT into ours, and completely
ifdeffed our own MDCT code out if a 32-bit fixed point MDCT was selected.
This is also the status quo nowadays - 2 separate MDCTs, one which
produces floating point and 16-bit fixed point versions, and one
sort-of integrated which produces 32-bit MDCT.
MIPS weren't all that interested in encoding, so they left the encoder
as-is, and they didn't care much about the ifdeffery, mess or quality - it's
not their problem.
So the MDCT/FFT code has always been a thorn in anyone looking to clean up
code's eye.
Backstory over. Internally AC3 operates on 25-bit fixed-point coefficients.
So for the floating point version, the encoder simply runs the float MDCT,
and converts the resulting coefficients to 25-bit fixed-point, as AC3 is inherently
a fixed-point codec. For the fixed-point version, the input is 16-bit samples,
so to maximize precision the frame samples are analyzed and the highest set
bit is detected via ac3_max_msb_abs_int16(), and the coefficients are then
scaled up via ac3_lshift_int16(), so the input for the FFT is always at least 14 bits,
computed in normalize_samples(). After FFT, the coefficients are scaled up to 25 bits.
This patch simply changes the encoder to accept 32-bit samples, reusing
the already well-optimized 32-bit MDCT code, allowing us to clean up and drop
a large part of a very messy code of ours, as well as prepare for the future lavu/tx
conversion. The coefficients are simply scaled down to 25 bits during windowing,
skipping 2 separate scalings, as the hacks to extend precision are simply no longer
necessary. There's no point in running the MDCT always at 32 bits when you're
going to drop 6 bits off anyway, the headroom is plenty, and the MDCT rounds
properly.
This also makes the encoder even slightly more accurate over the float version,
as there's no coefficient conversion step necessary.
SIZE SAVINGS:
ARM32:
HARDCODED TABLES:
BASE - 10709590
DROP DSP - 10702872 - diff: -6.56KiB
DROP MDCT - 10667932 - diff: -34.12KiB - both: -40.68KiB
DROP FFT - 10336652 - diff: -323.52KiB - all: -364.20KiB
SOFTCODED TABLES:
BASE - 9685096
DROP DSP - 9678378 - diff: -6.56KiB
DROP MDCT - 9643466 - diff: -34.09KiB - both: -40.65KiB
DROP FFT - 9573918 - diff: -67.92KiB - all: -108.57KiB
ARM64:
HARDCODED TABLES:
BASE - 14641112
DROP DSP - 14633806 - diff: -7.13KiB
DROP MDCT - 14604812 - diff: -28.31KiB - both: -35.45KiB
DROP FFT - 14286826 - diff: -310.53KiB - all: -345.98KiB
SOFTCODED TABLES:
BASE - 13636238
DROP DSP - 13628932 - diff: -7.13KiB
DROP MDCT - 13599866 - diff: -28.38KiB - both: -35.52KiB
DROP FFT - 13542080 - diff: -56.43KiB - all: -91.95KiB
x86:
HARDCODED TABLES:
BASE - 12367336
DROP DSP - 12354698 - diff: -12.34KiB
DROP MDCT - 12331024 - diff: -23.12KiB - both: -35.46KiB
DROP FFT - 12029788 - diff: -294.18KiB - all: -329.64KiB
SOFTCODED TABLES:
BASE - 11358094
DROP DSP - 11345456 - diff: -12.34KiB
DROP MDCT - 11321742 - diff: -23.16KiB - both: -35.50KiB
DROP FFT - 11276946 - diff: -43.75KiB - all: -79.25KiB
PERFORMANCE (10min random s32le):
ARM32 - before - 39.9x - 0m15.046s
ARM32 - after - 28.2x - 0m21.525s
Speed: -30%
ARM64 - before - 36.1x - 0m16.637s
ARM64 - after - 36.0x - 0m16.727s
Speed: -0.5%
x86 - before - 184x - 0m3.277s
x86 - after - 190x - 0m3.187s
Speed: +3%
2021-01-09 02:51:52 +02:00
|
|
|
#if AC3ENC_FLOAT
|
|
|
|
scale_coefficients(s);
|
|
|
|
#endif
|
2011-06-27 21:29:33 +03:00
|
|
|
|
2020-12-03 16:28:35 +02:00
|
|
|
return ff_ac3_encode_frame_common_end(avctx, avpkt, frame, got_packet_ptr);
|
2011-06-27 06:58:19 +03:00
|
|
|
}
|