From patchwork Tue Jan 12 18:12:10 2021
ContentType: text/plain; charset="utf8"
MIMEVersion: 1.0
ContentTransferEncoding: 8bit
XPatchworkSubmitter: Lynne
XPatchworkId: 24925
ReturnPath:
XOriginalTo: patchwork@ffauxbg.ffmpeg.org
DeliveredTo: patchwork@ffauxbg.ffmpeg.org
Received: from ffbox0bg.mplayerhq.hu (ffbox0bg.ffmpeg.org [79.124.17.100])
by ffaux.localdomain (Postfix) with ESMTP id 17FB844A3C0
for ; Tue, 12 Jan 2021 20:12:18 +0200 (EET)
Received: from [127.0.1.1] (localhost [127.0.0.1])
by ffbox0bg.mplayerhq.hu (Postfix) with ESMTP id E300168A94D;
Tue, 12 Jan 2021 20:12:17 +0200 (EET)
XOriginalTo: ffmpegdevel@ffmpeg.org
DeliveredTo: ffmpegdevel@ffmpeg.org
Received: from w4.tutanota.de (w4.tutanota.de [81.3.6.165])
by ffbox0bg.mplayerhq.hu (Postfix) with ESMTPS id C1BCE68A94D
for ; Tue, 12 Jan 2021 20:12:11 +0200 (EET)
Received: from w3.tutanota.de (unknown [192.168.1.164])
by w4.tutanota.de (Postfix) with ESMTP id 1C2271060308
for ; Tue, 12 Jan 2021 18:12:11 +0000 (UTC)
DKIMSignature: v=1; a=rsasha256; q=dns/txt; c=relaxed/relaxed; t=1610475130;
s=s1; d=lynne.ee;
h=From:From:To:To:Subject:Subject:ContentDescription:ContentID:ContentType:ContentType:ContentTransferEncoding:Cc:Date:Date:InReplyTo:MIMEVersion:MIMEVersion:MessageID:MessageID:ReplyTo:References:Sender;
bh=33mrJ+0cEZo62bc5B8jWfSVrZa8JtpptUTw7nJmh+xc=;
b=DQq/ycM6Dyad8rInGbBBYEQJx+7sUBzWDf1TF5HFaUvd8rxj4rQMCQGeH+xiCkWu
/0T3HPfyGcbIHSvQ7Y6V5wAB28+Jf9kngf6HzGwhvHQxEM8nT7H8eXZuxjs44diUFZe
+Z252YYM8ZJl5ZKpD5LbREfPRl4aQhdwa1x/lSf4SRPKK1JzKHLKEwrcHslWf4x+9E6
qwSg23/k4U+3HaNIKIcftPzaetaQAQS5Ff0yIMuvFpysow4zU9OoHc86RGKIqTjBcN1
Rc22tAAZ69KaCq/K9N9jpX42BturNcx3xVy2C8+yKANFg0hX8PrbINzLsWk+X/IBaI5
ePDb/NAIZw==
Date: Tue, 12 Jan 2021 19:12:10 +0100 (CET)
From: Lynne
To: Ffmpeg Devel
MessageID:
MIMEVersion: 1.0
Subject: [FFmpegdevel] [PATCH v2 1/5] ac3enc_fixed: convert to 32bit
sample format
XBeenThere: ffmpegdevel@ffmpeg.org
XMailmanVersion: 2.1.20
Precedence: list
ListId: FFmpeg development discussions and patches
ListUnsubscribe: ,
ListArchive:
ListPost:
ListHelp:
ListSubscribe: ,
ReplyTo: FFmpeg development discussions and patches
ErrorsTo: ffmpegdevelbounces@ffmpeg.org
Sender: "ffmpegdevel"
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 floatingpoint DSP was
simply replaced with FFmpeg's own functions.
However, FFmpeg had no fixedpoint audio code at that point. So
the encoder brought along its own fixedpoint DSP functions,
including a fixedpoint MDCT.
The fixedpoint MDCT itself is trivially just a float MDCT with a
different type and each multiply being a fixedpoint 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 fixedpoint version of the
encoder operates at 16bits 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 fixedpoint decoding on their FPUless cores. So they contributed
patches to template the AC3 decoder so it had both a fixedpoint
and a floatingpoint version. They also did the same for the AAC decoder.
They however, used 32bit samples. Not 16bits. And we did not have
32bit fixedpoint DSP functions, including an MDCT. But instead of
templating our MDCT to output 3 versions (float, 32bit fixed and 16bit fixed),
they simply copypasted their own MDCT into ours, and completely
ifdeffed our own MDCT code out if a 32bit fixed point MDCT was selected.
This is also the status quo nowadays  2 separate MDCTs, one which
produces floating point and 16bit fixed point versions, and one
sortof integrated which produces 32bit MDCT.
MIPS weren't all that interested in encoding, so they left the encoder
asis, 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 25bit fixedpoint coefficients.
So for the floating point version, the encoder simply runs the float MDCT,
and converts the resulting coefficients to 25bit fixedpoint, as AC3 is inherently
a fixedpoint codec. For the fixedpoint version, the input is 16bit 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 32bit samples, reusing
the already welloptimized 32bit 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%
New patch attached.
Subject: [PATCH v2 1/5] ac3enc_fixed: convert to 32bit 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 floatingpoint DSP was
simply replaced with FFmpeg's own functions.
However, FFmpeg had no fixedpoint audio code at that point. So
the encoder brought along its own fixedpoint DSP functions,
including a fixedpoint MDCT.
The fixedpoint MDCT itself is trivially just a float MDCT with a
different type and each multiply being a fixedpoint 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 fixedpoint version of the
encoder operates at 16bits 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 fixedpoint decoding on their FPUless cores. So they contributed
patches to template the AC3 decoder so it had both a fixedpoint
and a floatingpoint version. They also did the same for the AAC decoder.
They however, used 32bit samples. Not 16bits. And we did not have
32bit fixedpoint DSP functions, including an MDCT. But instead of
templating our MDCT to output 3 versions (float, 32bit fixed and 16bit fixed),
they simply copypasted their own MDCT into ours, and completely
ifdeffed our own MDCT code out if a 32bit fixed point MDCT was selected.
This is also the status quo nowadays  2 separate MDCTs, one which
produces floating point and 16bit fixed point versions, and one
sortof integrated which produces 32bit MDCT.
MIPS weren't all that interested in encoding, so they left the encoder
asis, 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 25bit fixedpoint coefficients.
So for the floating point version, the encoder simply runs the float MDCT,
and converts the resulting coefficients to 25bit fixedpoint, as AC3 is inherently
a fixedpoint codec. For the fixedpoint version, the input is 16bit 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 32bit samples, reusing
the already welloptimized 32bit 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%

doc/encoders.texi  7 ++
libavcodec/Makefile  2 +
libavcodec/ac3enc.c  1 +
libavcodec/ac3enc.h  11 +++
libavcodec/ac3enc_fixed.c  59 ++++++++++++
libavcodec/ac3enc_float.c  1 
libavcodec/ac3enc_template.c  21 ++++
libavcodec/version.h  2 +
tests/fate/ac3.mak  2 +
tests/fate/ffmpeg.mak  2 +
tests/ref/fate/unknown_layoutac3  2 +
tests/ref/lavf/rm  2 +
12 files changed, 45 insertions(+), 67 deletions()
diff git a/doc/encoders.texi b/doc/encoders.texi
index 0b1c69e982..60e763a704 100644
 a/doc/encoders.texi
+++ b/doc/encoders.texi
@@ 151,10 +151,9 @@ the undocumented RealAudio 3 (a.k.a. dnet).
The @var{ac3} encoder uses floatingpoint math, while the @var{ac3_fixed}
encoder only uses fixedpoint integer math. This does not mean that one is
always faster, just that one or the other may be better suited to a
particular system. The floatingpoint encoder will generally produce better
quality audio for a given bitrate. The @var{ac3_fixed} encoder is not the
default codec for any of the output formats, so it must be specified explicitly
using the option @code{acodec ac3_fixed} in order to use it.
+particular system. The @var{ac3_fixed} encoder is not the default codec for
+any of the output formats, so it must be specified explicitly using the option
+@code{acodec ac3_fixed} in order to use it.
@subsection AC3 Metadata
diff git a/libavcodec/Makefile b/libavcodec/Makefile
index 35318f4f4d..0546e6f6c5 100644
 a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ 181,7 +181,7 @@ OBJS$(CONFIG_AC3_DECODER) += ac3dec_float.o ac3dec_data.o ac3.o kbd
OBJS$(CONFIG_AC3_FIXED_DECODER) += ac3dec_fixed.o ac3dec_data.o ac3.o kbdwin.o ac3tab.o
OBJS$(CONFIG_AC3_ENCODER) += ac3enc_float.o ac3enc.o ac3tab.o \
ac3.o kbdwin.o
OBJS$(CONFIG_AC3_FIXED_ENCODER) += ac3enc_fixed.o ac3enc.o ac3tab.o ac3.o
+OBJS$(CONFIG_AC3_FIXED_ENCODER) += ac3enc_fixed.o ac3enc.o ac3tab.o ac3.o kbdwin.o
OBJS$(CONFIG_AC3_MF_ENCODER) += mfenc.o mf_utils.o
OBJS$(CONFIG_ACELP_KELVIN_DECODER) += g729dec.o lsp.o celp_math.o celp_filters.o acelp_filters.o acelp_pitch_delay.o acelp_vectors.o g729postfilter.o
OBJS$(CONFIG_AGM_DECODER) += agm.o
diff git a/libavcodec/ac3enc.c b/libavcodec/ac3enc.c
index b2e3b2bb4b..9dafe0ef55 100644
 a/libavcodec/ac3enc.c
+++ b/libavcodec/ac3enc.c
@@ 2047,6 +2047,7 @@ av_cold int ff_ac3_encode_close(AVCodecContext *avctx)
int blk, ch;
AC3EncodeContext *s = avctx>priv_data;
+ av_freep(&s>mdct_window);
av_freep(&s>windowed_samples);
if (s>planar_samples)
for (ch = 0; ch < s>channels; ch++)
diff git a/libavcodec/ac3enc.h b/libavcodec/ac3enc.h
index 044564ecb4..ba62891371 100644
 a/libavcodec/ac3enc.h
+++ b/libavcodec/ac3enc.h
@@ 30,8 +30,6 @@
#include
#include "libavutil/float_dsp.h"

#include "ac3.h"
#include "ac3dsp.h"
#include "avcodec.h"
@@ 53,6 +51,7 @@
#define AC3ENC_TYPE_EAC3 2
#if AC3ENC_FLOAT
+#include "libavutil/float_dsp.h"
#define AC3_NAME(x) ff_ac3_float_ ## x
#define MAC_COEF(d,a,b) ((d)+=(a)*(b))
#define COEF_MIN (16777215.0/16777216.0)
@@ 62,12 +61,13 @@ typedef float SampleType;
typedef float CoefType;
typedef float CoefSumType;
#else
+#include "libavutil/fixed_dsp.h"
#define AC3_NAME(x) ff_ac3_fixed_ ## x
#define MAC_COEF(d,a,b) MAC64(d,a,b)
#define COEF_MIN 16777215
#define COEF_MAX 16777215
#define NEW_CPL_COORD_THRESHOLD 503317
typedef int16_t SampleType;
+typedef int32_t SampleType;
typedef int32_t CoefType;
typedef int64_t CoefSumType;
#endif
@@ 141,7 +141,6 @@ typedef struct AC3Block {
uint16_t **qmant; ///< quantized mantissas
uint8_t **cpl_coord_exp; ///< coupling coord exponents (cplcoexp)
uint8_t **cpl_coord_mant; ///< coupling coord mantissas (cplcomant)
 uint8_t coeff_shift[AC3_MAX_CHANNELS]; ///< fixedpoint coefficient shift values
uint8_t new_rematrixing_strategy; ///< send new rematrixing flags in this block
int num_rematrixing_bands; ///< number of rematrixing bands
uint8_t rematrixing_flags[4]; ///< rematrixing flags
@@ 165,7 +164,11 @@ typedef struct AC3EncodeContext {
AVCodecContext *avctx; ///< parent AVCodecContext
PutBitContext pb; ///< bitstream writer context
AudioDSPContext adsp;
+#if AC3ENC_FLOAT
AVFloatDSPContext *fdsp;
+#else
+ AVFixedDSPContext *fdsp;
+#endif
MECmpContext mecc;
AC3DSPContext ac3dsp; ///< AC3 optimized functions
FFTContext mdct; ///< FFT context for MDCT calculation
diff git a/libavcodec/ac3enc_fixed.c b/libavcodec/ac3enc_fixed.c
index 7818dd8c35..3b302d40df 100644
 a/libavcodec/ac3enc_fixed.c
+++ b/libavcodec/ac3enc_fixed.c
@@ 26,12 +26,14 @@
* fixedpoint AC3 encoder.
*/
#define FFT_FLOAT 0
#define AC3ENC_FLOAT 0
+#define FFT_FLOAT 0
+#define FFT_FIXED_32 1
#include "internal.h"
#include "audiodsp.h"
#include "ac3enc.h"
#include "eac3enc.h"
+#include "kbdwin.h"
#define AC3ENC_TYPE AC3ENC_TYPE_AC3_FIXED
#include "ac3enc_opts_template.c"
@@ 43,37 +45,6 @@ static const AVClass ac3enc_class = {
.version = LIBAVUTIL_VERSION_INT,
};
/*
 * Normalize the input samples to use the maximum available precision.
 * This assumes signed 16bit input samples.
 */
static int normalize_samples(AC3EncodeContext *s)
{
 int v = s>ac3dsp.ac3_max_msb_abs_int16(s>windowed_samples, AC3_WINDOW_SIZE);
 v = 14  av_log2(v);
 if (v > 0)
 s>ac3dsp.ac3_lshift_int16(s>windowed_samples, AC3_WINDOW_SIZE, v);
 /* +6 to rightshift from 31bit to 25bit */
 return v + 6;
}


/*
 * Scale MDCT coefficients to 25bit signed fixedpoint.
 */
static void scale_coefficients(AC3EncodeContext *s)
{
 int blk, ch;

 for (blk = 0; blk < s>num_blocks; blk++) {
 AC3Block *block = &s>blocks[blk];
 for (ch = 1; ch <= s>channels; ch++) {
 s>ac3dsp.ac3_rshift_int32(block>mdct_coef[ch], AC3_MAX_COEFS,
 block>coeff_shift[ch]);
 }
 }
}

static void sum_square_butterfly(AC3EncodeContext *s, int64_t sum[4],
const int32_t *coef0, const int32_t *coef1,
int len)
@@ 120,7 +91,6 @@ static av_cold void ac3_fixed_mdct_end(AC3EncodeContext *s)
ff_mdct_end(&s>mdct);
}

/**
* Initialize MDCT tables.
*
@@ 129,9 +99,24 @@ static av_cold void ac3_fixed_mdct_end(AC3EncodeContext *s)
*/
static av_cold int ac3_fixed_mdct_init(AC3EncodeContext *s)
{
 int ret = ff_mdct_init(&s>mdct, 9, 0, 1.0);
 s>mdct_window = ff_ac3_window;
 return ret;
+ float fwin[AC3_BLOCK_SIZE];
+
+ int32_t *iwin = av_malloc_array(AC3_WINDOW_SIZE, sizeof(*iwin));
+ if (!iwin)
+ return AVERROR(ENOMEM);
+
+ ff_kbd_window_init(fwin, 5.0, AC3_WINDOW_SIZE/2);
+
+ for (int i = 0; i < AC3_WINDOW_SIZE/2; i++)
+ iwin[AC3_WINDOW_SIZE1i] = lrintf(fwin[i] * (1 << 22));
+
+ s>mdct_window = iwin;
+
+ s>fdsp = avpriv_alloc_fixed_dsp(s>avctx>flags & AV_CODEC_FLAG_BITEXACT);
+ if (!s>fdsp)
+ return AVERROR(ENOMEM);
+
+ return ff_mdct_init(&s>mdct, 9, 0, 1.0);
}
@@ 155,7 +140,7 @@ AVCodec ff_ac3_fixed_encoder = {
.init = ac3_fixed_encode_init,
.encode2 = ff_ac3_fixed_encode_frame,
.close = ff_ac3_encode_close,
 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16P,
+ .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32P,
AV_SAMPLE_FMT_NONE },
.priv_class = &ac3enc_class,
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE  FF_CODEC_CAP_INIT_CLEANUP,
diff git a/libavcodec/ac3enc_float.c b/libavcodec/ac3enc_float.c
index 45bfed34f9..b17b3a2365 100644
 a/libavcodec/ac3enc_float.c
+++ b/libavcodec/ac3enc_float.c
@@ 97,7 +97,6 @@ static void sum_square_butterfly(AC3EncodeContext *s, float sum[4],
static av_cold void ac3_float_mdct_end(AC3EncodeContext *s)
{
ff_mdct_end(&s>mdct);
 av_freep(&s>mdct_window);
}
diff git a/libavcodec/ac3enc_template.c b/libavcodec/ac3enc_template.c
index 0fdc95b968..de6eba71d8 100644
 a/libavcodec/ac3enc_template.c
+++ b/libavcodec/ac3enc_template.c
@@ 91,18 +91,11 @@ static void apply_mdct(AC3EncodeContext *s)
AC3Block *block = &s>blocks[blk];
const SampleType *input_samples = &s>planar_samples[ch][blk * AC3_BLOCK_SIZE];
#if AC3ENC_FLOAT
s>fdsp>vector_fmul(s>windowed_samples, input_samples,
 s>mdct_window, AC3_WINDOW_SIZE);
#else
 s>ac3dsp.apply_window_int16(s>windowed_samples, input_samples,
 s>mdct_window, AC3_WINDOW_SIZE);

 block>coeff_shift[ch + 1] = normalize_samples(s);
#endif
+ s>mdct_window, AC3_WINDOW_SIZE);
 s>mdct.mdct_calcw(&s>mdct, block>mdct_coef[ch+1],
 s>windowed_samples);
+ s>mdct.mdct_calc(&s>mdct, block>mdct_coef[ch+1],
+ s>windowed_samples);
}
}
}
@@ 390,9 +383,6 @@ int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
apply_mdct(s);
 if (!AC3ENC_FLOAT)
 scale_coefficients(s);

clip_coefficients(&s>adsp, s>blocks[0].mdct_coef[1],
AC3_MAX_COEFS * s>num_blocks * s>channels);
@@ 404,8 +394,9 @@ int AC3_NAME(encode_frame)(AVCodecContext *avctx, AVPacket *avpkt,
compute_rematrixing_strategy(s);
 if (AC3ENC_FLOAT)
 scale_coefficients(s);
+#if AC3ENC_FLOAT
+ scale_coefficients(s);
+#endif
return ff_ac3_encode_frame_common_end(avctx, avpkt, frame, got_packet_ptr);
}
diff git a/libavcodec/version.h b/libavcodec/version.h
index 1420439044..cd871f0fa0 100644
 a/libavcodec/version.h
+++ b/libavcodec/version.h
@@ 28,7 +28,7 @@
#include "libavutil/version.h"
#define LIBAVCODEC_VERSION_MAJOR 58
#define LIBAVCODEC_VERSION_MINOR 116
+#define LIBAVCODEC_VERSION_MINOR 117
#define LIBAVCODEC_VERSION_MICRO 100
#define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
diff git a/tests/fate/ac3.mak b/tests/fate/ac3.mak
index 757cd51cf2..d76e22bade 100644
 a/tests/fate/ac3.mak
+++ b/tests/fate/ac3.mak
@@ 90,7 +90,7 @@ fateac3fixedencode: tests/data/asynth441002.wav
fateac3fixedencode: SRC = $(TARGET_PATH)/tests/data/asynth441002.wav
fateac3fixedencode: CMD = md5 i $(SRC) c ac3_fixed ab 128k f ac3 flags +bitexact af aresample
fateac3fixedencode: CMP = oneline
fateac3fixedencode: REF = a1d1fc116463b771abf5aef7ed37d7b1
+fateac3fixedencode: REF = 1f548175e11a95e62ce20e442fcc8d08
FATE_EAC3$(call ALLYES, EAC3_DEMUXER EAC3_MUXER EAC3_CORE_BSF) += fateeac3corebsf
fateeac3corebsf: CMD = md5pipe i $(TARGET_SAMPLES)/eac3/the_great_wall_7.1.eac3 c:a copy bsf:a eac3_core fflags +bitexact f eac3
diff git a/tests/fate/ffmpeg.mak b/tests/fate/ffmpeg.mak
index c6d8dc2e5c..4dfb77d250 100644
 a/tests/fate/ffmpeg.mak
+++ b/tests/fate/ffmpeg.mak
@@ 83,7 +83,7 @@ fateunknown_layoutpcm: CMD = md5 \
FATE_FFMPEG$(call ALLYES, PCM_S16LE_DEMUXER AC3_MUXER PCM_S16LE_DECODER AC3_FIXED_ENCODER) += fateunknown_layoutac3
fateunknown_layoutac3: $(AREF)
fateunknown_layoutac3: CMD = md5 auto_conversion_filters \
 guess_layout_max 0 f s16le ac 1 ar 44100 i $(TARGET_PATH)/$(AREF) \
+ guess_layout_max 0 f s32le ac 1 ar 44100 i $(TARGET_PATH)/$(AREF) \
f ac3 flags +bitexact c ac3_fixed
diff git a/tests/ref/fate/unknown_layoutac3 b/tests/ref/fate/unknown_layoutac3
index d332efcec4..719a44aacf 100644
 a/tests/ref/fate/unknown_layoutac3
+++ b/tests/ref/fate/unknown_layoutac3
@@ 1 +1 @@
bbb7550d6d93973c10f4ee13c87cf799
+febdb165cfd6cba375aa086195e61213
diff git a/tests/ref/lavf/rm b/tests/ref/lavf/rm
index 43ea4c7897..fc2a6564a2 100644
 a/tests/ref/lavf/rm
+++ b/tests/ref/lavf/rm
@@ 1,2 +1,2 @@
e30681d05d6f3d24108d3614600bf116 *tests/data/lavf/lavf.rm
+8dfb8d4556d61d3615e0d0012ffe540c *tests/data/lavf/lavf.rm
346424 tests/data/lavf/lavf.rm