* [FFmpeg-devel] [PATCH] aacdec: convert to lavu/tx and support fixed-point 960-sample decoding
@ 2022-10-28 21:46 Lynne
2022-10-30 19:39 ` Michael Niedermayer
0 siblings, 1 reply; 2+ messages in thread
From: Lynne @ 2022-10-28 21:46 UTC (permalink / raw)
To: Ffmpeg Devel
[-- Attachment #1: Type: text/plain, Size: 582 bytes --]
This patch replaces the transform used in AAC with lavu/tx and removes
the limitation on only being able to decode 960-sample files
with the float decoder.
This commit also removes a whole bunch of unnecessary and slow
lifting steps the decoder did to compensate for the poor accuracy
of the old integer transformation code.
Overall float decoder speedup on Zen 3 for 64kbps: 32%
Patch attached.
Note: lavu/tx's PFA 960pt transform currently has a bug that'll get fixed
by my next patchset. The naive transforms are unaffected and may
be used to test 960pt fixed-point decoding.
[-- Attachment #2: 0001-aacdec-convert-to-lavu-tx-and-support-fixed-point-96.patch --]
[-- Type: text/x-diff, Size: 16852 bytes --]
From 95625b470fb61676e838310dc0108eb648f972a3 Mon Sep 17 00:00:00 2001
From: Lynne <dev@lynne.ee>
Date: Fri, 28 Oct 2022 21:36:10 +0200
Subject: [PATCH] aacdec: convert to lavu/tx and support fixed-point 960-sample
decoding
This patch replaces the transform used in AAC with lavu/tx and removes
the limitation on only being able to decode 960-sample files
with the float decoder.
This commit also removes a whole bunch of unnecessary and slow
lifting steps the decoder did to compensate for the poor accuracy
of the old integer transformation code.
Overall float decoder speedup on Zen 3 for 64kbps: 32%
---
configure | 4 +-
libavcodec/aac.h | 27 ++++---
libavcodec/aac_defines.h | 4 +-
libavcodec/aacdec.c | 4 +-
libavcodec/aacdec_fixed.c | 5 +-
libavcodec/aacdec_template.c | 106 ++++++++++------------------
libavcodec/aacsbr.c | 1 -
libavcodec/aacsbr_fixed.c | 1 -
libavcodec/mips/aacdec_mips.c | 4 +-
libavcodec/sinewin_fixed_tablegen.c | 3 +
libavcodec/sinewin_fixed_tablegen.h | 6 ++
11 files changed, 74 insertions(+), 91 deletions(-)
diff --git a/configure b/configure
index 70c9e41dcc..97a7cdd7fd 100755
--- a/configure
+++ b/configure
@@ -2763,8 +2763,8 @@ vc1dsp_select="h264chroma qpeldsp startcode"
rdft_select="fft"
# decoders / encoders
-aac_decoder_select="adts_header mdct15 mdct mpeg4audio sinewin"
-aac_fixed_decoder_select="adts_header mdct mpeg4audio"
+aac_decoder_select="adts_header mpeg4audio sinewin"
+aac_fixed_decoder_select="adts_header mpeg4audio"
aac_encoder_select="audio_frame_queue iirfilter lpc mdct sinewin"
aac_latm_decoder_select="aac_decoder aac_latm_parser"
ac3_decoder_select="ac3_parser ac3dsp bswapdsp fmtconvert mdct"
diff --git a/libavcodec/aac.h b/libavcodec/aac.h
index c8d6b17710..285d3b7482 100644
--- a/libavcodec/aac.h
+++ b/libavcodec/aac.h
@@ -36,11 +36,8 @@
#include "libavutil/float_dsp.h"
#include "libavutil/fixed_dsp.h"
#include "libavutil/mem_internal.h"
+#include "libavutil/tx.h"
#include "avcodec.h"
-#if !USE_FIXED
-#include "mdct15.h"
-#endif
-#include "fft.h"
#include "mpeg4audio.h"
#include "sbr.h"
@@ -326,16 +323,24 @@ struct AACContext {
* @name Computed / set up during initialization
* @{
*/
- FFTContext mdct;
- FFTContext mdct_small;
- FFTContext mdct_ld;
- FFTContext mdct_ltp;
+ AVTXContext *mdct120;
+ AVTXContext *mdct128;
+ AVTXContext *mdct480;
+ AVTXContext *mdct512;
+ AVTXContext *mdct960;
+ AVTXContext *mdct1024;
+ AVTXContext *mdct_ltp;
+
+ av_tx_fn mdct120_fn;
+ av_tx_fn mdct128_fn;
+ av_tx_fn mdct480_fn;
+ av_tx_fn mdct512_fn;
+ av_tx_fn mdct960_fn;
+ av_tx_fn mdct1024_fn;
+ av_tx_fn mdct_ltp_fn;
#if USE_FIXED
AVFixedDSPContext *fdsp;
#else
- MDCT15Context *mdct120;
- MDCT15Context *mdct480;
- MDCT15Context *mdct960;
AVFloatDSPContext *fdsp;
#endif /* USE_FIXED */
int random_state;
diff --git a/libavcodec/aac_defines.h b/libavcodec/aac_defines.h
index a3c680b65c..21063c4d58 100644
--- a/libavcodec/aac_defines.h
+++ b/libavcodec/aac_defines.h
@@ -45,7 +45,7 @@ typedef int AAC_SIGNE;
#define Q23(a) (int)((a) * 8388608.0 + 0.5)
#define Q30(x) (int)((x)*1073741824.0 + 0.5)
#define Q31(x) (int)((x)*2147483648.0 + 0.5)
-#define RANGE15(x) x
+#define TX_SCALE(x) ((x) * 128.0f)
#define GET_GAIN(x, y) (-(y) * (1 << (x))) + 1024
#define AAC_MUL16(x, y) (int)(((int64_t)(x) * (y) + 0x8000) >> 16)
#define AAC_MUL26(x, y) (int)(((int64_t)(x) * (y) + 0x2000000) >> 26)
@@ -94,7 +94,7 @@ typedef unsigned AAC_SIGNE;
#define Q23(x) ((float)(x))
#define Q30(x) ((float)(x))
#define Q31(x) ((float)(x))
-#define RANGE15(x) (32768.0 * (x))
+#define TX_SCALE(x) ((x) / 32768.0f)
#define GET_GAIN(x, y) powf((x), -(y))
#define AAC_MUL16(x, y) ((x) * (y))
#define AAC_MUL26(x, y) ((x) * (y))
diff --git a/libavcodec/aacdec.c b/libavcodec/aacdec.c
index 2fdfd6b221..ca31540d3c 100644
--- a/libavcodec/aacdec.c
+++ b/libavcodec/aacdec.c
@@ -32,16 +32,14 @@
* @author Maxim Gavrilov ( maxim.gavrilov gmail com )
*/
-#define FFT_FLOAT 1
#define USE_FIXED 0
+#define TX_TYPE AV_TX_FLOAT_MDCT
#include "libavutil/float_dsp.h"
#include "libavutil/opt.h"
#include "avcodec.h"
#include "codec_internal.h"
#include "get_bits.h"
-#include "fft.h"
-#include "mdct15.h"
#include "lpc.h"
#include "kbdwin.h"
#include "sinewin.h"
diff --git a/libavcodec/aacdec_fixed.c b/libavcodec/aacdec_fixed.c
index 8c5dad2813..46f0ebf91b 100644
--- a/libavcodec/aacdec_fixed.c
+++ b/libavcodec/aacdec_fixed.c
@@ -58,15 +58,14 @@
* @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
*/
-#define FFT_FLOAT 0
#define USE_FIXED 1
+#define TX_TYPE AV_TX_INT32_MDCT
#include "libavutil/fixed_dsp.h"
#include "libavutil/opt.h"
#include "avcodec.h"
#include "codec_internal.h"
#include "get_bits.h"
-#include "fft.h"
#include "lpc.h"
#include "kbdwin.h"
#include "sinewin_fixed_tablegen.h"
@@ -87,6 +86,8 @@
DECLARE_ALIGNED(32, static int, AAC_RENAME2(aac_kbd_long_1024))[1024];
DECLARE_ALIGNED(32, static int, AAC_RENAME2(aac_kbd_short_128))[128];
+DECLARE_ALIGNED(32, static int, AAC_RENAME2(aac_kbd_long_960))[960];
+DECLARE_ALIGNED(32, static int, AAC_RENAME2(aac_kbd_short_120))[120];
static av_always_inline void reset_predict_state(PredictorState *ps)
{
diff --git a/libavcodec/aacdec_template.c b/libavcodec/aacdec_template.c
index 4a9513d53e..cdf266dbfa 100644
--- a/libavcodec/aacdec_template.c
+++ b/libavcodec/aacdec_template.c
@@ -931,13 +931,6 @@ static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
uint8_t layout_map[MAX_ELEM_ID*4][3];
int tags = 0;
-#if USE_FIXED
- if (get_bits1(gb)) { // frameLengthFlag
- avpriv_report_missing_feature(avctx, "Fixed point 960/120 MDCT window");
- return AVERROR_PATCHWELCOME;
- }
- m4ac->frame_length_short = 0;
-#else
m4ac->frame_length_short = get_bits1(gb);
if (m4ac->frame_length_short && m4ac->sbr == 1) {
avpriv_report_missing_feature(avctx, "SBR with 960 frame length");
@@ -945,7 +938,6 @@ static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
m4ac->sbr = 0;
m4ac->ps = 0;
}
-#endif
if (get_bits1(gb)) // dependsOnCoreCoder
skip_bits(gb, 14); // coreCoderDelay
@@ -1022,14 +1014,8 @@ static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
m4ac->ps = 0;
m4ac->sbr = 0;
-#if USE_FIXED
- if (get_bits1(gb)) { // frameLengthFlag
- avpriv_request_sample(avctx, "960/120 MDCT window");
- return AVERROR_PATCHWELCOME;
- }
-#else
m4ac->frame_length_short = get_bits1(gb);
-#endif
+
res_flags = get_bits(gb, 3);
if (res_flags) {
avpriv_report_missing_feature(avctx,
@@ -1256,9 +1242,10 @@ static av_cold void aac_static_table_init(void)
352);
// window initialization
-#if !USE_FIXED
AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(aac_kbd_long_960), 4.0, 960);
AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(aac_kbd_short_120), 6.0, 120);
+
+#if !USE_FIXED
AAC_RENAME(ff_sine_window_init)(AAC_RENAME(sine_960), 960);
AAC_RENAME(ff_sine_window_init)(AAC_RENAME(sine_120), 120);
AAC_RENAME(ff_init_ff_sine_windows)(9);
@@ -1276,6 +1263,7 @@ static AVOnce aac_table_init = AV_ONCE_INIT;
static av_cold int aac_decode_init(AVCodecContext *avctx)
{
+ float scale;
AACContext *ac = avctx->priv_data;
int ret;
@@ -1348,21 +1336,25 @@ static av_cold int aac_decode_init(AVCodecContext *avctx)
ac->random_state = 0x1f2e3d4c;
- AAC_RENAME_32(ff_mdct_init)(&ac->mdct, 11, 1, 1.0 / RANGE15(1024.0));
- AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ld, 10, 1, 1.0 / RANGE15(512.0));
- AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
- AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
-#if !USE_FIXED
- ret = ff_mdct15_init(&ac->mdct120, 1, 3, 1.0f/(16*1024*120*2));
- if (ret < 0)
- return ret;
- ret = ff_mdct15_init(&ac->mdct480, 1, 5, 1.0f/(16*1024*960));
- if (ret < 0)
+#define MDCT_INIT(s, fn, len, sval) \
+ scale = sval; \
+ ret = av_tx_init(&s, &fn, TX_TYPE, 1, len, &scale, 0); \
+ if (ret < 0) \
return ret;
- ret = ff_mdct15_init(&ac->mdct960, 1, 6, 1.0f/(16*1024*960*2));
+
+ MDCT_INIT(ac->mdct120, ac->mdct120_fn, 120, TX_SCALE(1.0/120))
+ MDCT_INIT(ac->mdct128, ac->mdct128_fn, 128, TX_SCALE(1.0/128))
+ MDCT_INIT(ac->mdct480, ac->mdct480_fn, 480, TX_SCALE(1.0/480))
+ MDCT_INIT(ac->mdct512, ac->mdct512_fn, 512, TX_SCALE(1.0/512))
+ MDCT_INIT(ac->mdct960, ac->mdct960_fn, 960, TX_SCALE(1.0/960))
+ MDCT_INIT(ac->mdct1024, ac->mdct1024_fn, 1024, TX_SCALE(1.0/1024))
+#undef MDCT_INIT
+
+ /* LTP forward MDCT */
+ scale = USE_FIXED ? -1.0 : -32786.0*2 + 36;
+ ret = av_tx_init(&ac->mdct_ltp, &ac->mdct_ltp_fn, TX_TYPE, 0, 1024, &scale, 0);
if (ret < 0)
return ret;
-#endif
return 0;
}
@@ -2691,7 +2683,7 @@ static void windowing_and_mdct_ltp(AACContext *ac, INTFLOAT *out,
ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
memset(in + 1024 + 576, 0, 448 * sizeof(*in));
}
- ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
+ ac->mdct_ltp_fn(ac->mdct_ltp, out, in, sizeof(INTFLOAT));
}
/**
@@ -2783,13 +2775,9 @@ static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
// imdct
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
for (i = 0; i < 1024; i += 128)
- ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
+ ac->mdct128_fn(ac->mdct128, buf + i, in + i, sizeof(INTFLOAT));
} else {
- ac->mdct.imdct_half(&ac->mdct, buf, in);
-#if USE_FIXED
- for (i=0; i<1024; i++)
- buf[i] = (buf[i] + 4LL) >> 3;
-#endif /* USE_FIXED */
+ ac->mdct1024_fn(ac->mdct1024, buf, in, sizeof(INTFLOAT));
}
/* window overlapping
@@ -2837,7 +2825,6 @@ static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
*/
static void imdct_and_windowing_960(AACContext *ac, SingleChannelElement *sce)
{
-#if !USE_FIXED
IndividualChannelStream *ics = &sce->ics;
INTFLOAT *in = sce->coeffs;
INTFLOAT *out = sce->ret;
@@ -2852,9 +2839,9 @@ static void imdct_and_windowing_960(AACContext *ac, SingleChannelElement *sce)
// imdct
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
for (i = 0; i < 8; i++)
- ac->mdct120->imdct_half(ac->mdct120, buf + i * 120, in + i * 128, 1);
+ ac->mdct120_fn(ac->mdct120, buf + i * 120, in + i * 128, sizeof(INTFLOAT));
} else {
- ac->mdct960->imdct_half(ac->mdct960, buf, in, 1);
+ ac->mdct960_fn(ac->mdct960, buf, in, sizeof(INTFLOAT));
}
/* window overlapping
@@ -2896,7 +2883,6 @@ static void imdct_and_windowing_960(AACContext *ac, SingleChannelElement *sce)
} else { // LONG_STOP or ONLY_LONG
memcpy( saved, buf + 480, 480 * sizeof(*saved));
}
-#endif
}
static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
{
@@ -2905,17 +2891,9 @@ static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
INTFLOAT *out = sce->ret;
INTFLOAT *saved = sce->saved;
INTFLOAT *buf = ac->buf_mdct;
-#if USE_FIXED
- int i;
-#endif /* USE_FIXED */
// imdct
- ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
-
-#if USE_FIXED
- for (i = 0; i < 1024; i++)
- buf[i] = (buf[i] + 2) >> 2;
-#endif /* USE_FIXED */
+ ac->mdct512_fn(ac->mdct512, buf, in, sizeof(INTFLOAT));
// window overlapping
if (ics->use_kb_window[1]) {
@@ -2954,20 +2932,15 @@ static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
}
-#if !USE_FIXED
+
if (n == 480)
- ac->mdct480->imdct_half(ac->mdct480, buf, in, 1);
+ ac->mdct480_fn(ac->mdct480, buf, in, sizeof(INTFLOAT));
else
-#endif
- ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
-
-#if USE_FIXED
- for (i = 0; i < 1024; i++)
- buf[i] = (buf[i] + 1) >> 1;
-#endif /* USE_FIXED */
+ ac->mdct512_fn(ac->mdct512, buf, in, sizeof(INTFLOAT));
for (i = 0; i < n; i+=2) {
- buf[i] = -buf[i];
+ buf[i + 0] = -(USE_FIXED + 1)*buf[i + 0];
+ buf[i + 1] = (USE_FIXED + 1)*buf[i + 1];
}
// Like with the regular IMDCT at this point we still have the middle half
// of a transform but with even symmetry on the left and odd symmetry on
@@ -3529,15 +3502,14 @@ static av_cold int aac_decode_close(AVCodecContext *avctx)
}
}
- ff_mdct_end(&ac->mdct);
- ff_mdct_end(&ac->mdct_small);
- ff_mdct_end(&ac->mdct_ld);
- ff_mdct_end(&ac->mdct_ltp);
-#if !USE_FIXED
- ff_mdct15_uninit(&ac->mdct120);
- ff_mdct15_uninit(&ac->mdct480);
- ff_mdct15_uninit(&ac->mdct960);
-#endif
+ av_tx_uninit(&ac->mdct120);
+ av_tx_uninit(&ac->mdct128);
+ av_tx_uninit(&ac->mdct480);
+ av_tx_uninit(&ac->mdct512);
+ av_tx_uninit(&ac->mdct960);
+ av_tx_uninit(&ac->mdct1024);
+ av_tx_uninit(&ac->mdct_ltp);
+
av_freep(&ac->fdsp);
return 0;
}
diff --git a/libavcodec/aacsbr.c b/libavcodec/aacsbr.c
index b74cd82ac0..47aa6cb3c1 100644
--- a/libavcodec/aacsbr.c
+++ b/libavcodec/aacsbr.c
@@ -31,7 +31,6 @@
#include "sbr.h"
#include "aacsbr.h"
#include "aacsbrdata.h"
-#include "fft.h"
#include "internal.h"
#include "aacps.h"
#include "sbrdsp.h"
diff --git a/libavcodec/aacsbr_fixed.c b/libavcodec/aacsbr_fixed.c
index ffa63f2647..cac472552e 100644
--- a/libavcodec/aacsbr_fixed.c
+++ b/libavcodec/aacsbr_fixed.c
@@ -60,7 +60,6 @@
#include "sbr.h"
#include "aacsbr.h"
#include "aacsbrdata.h"
-#include "fft.h"
#include "aacps.h"
#include "sbrdsp.h"
#include "libavutil/internal.h"
diff --git a/libavcodec/mips/aacdec_mips.c b/libavcodec/mips/aacdec_mips.c
index 7b5f18a5d8..cd357cedbc 100644
--- a/libavcodec/mips/aacdec_mips.c
+++ b/libavcodec/mips/aacdec_mips.c
@@ -126,9 +126,9 @@ static void imdct_and_windowing_mips(AACContext *ac, SingleChannelElement *sce)
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
for (i = 0; i < 1024; i += 128)
- ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
+ ac->mdct128_fn(ac->mdct128, buf + i, in + i, sizeof(float));
} else
- ac->mdct.imdct_half(&ac->mdct, buf, in);
+ ac->mdct1024_fn(ac->mdct1024, buf, in, sizeof(float));
/* window overlapping
* NOTE: To simplify the overlapping code, all 'meaningless' short to long
diff --git a/libavcodec/sinewin_fixed_tablegen.c b/libavcodec/sinewin_fixed_tablegen.c
index 61e5274afa..15f0cc2072 100644
--- a/libavcodec/sinewin_fixed_tablegen.c
+++ b/libavcodec/sinewin_fixed_tablegen.c
@@ -35,8 +35,11 @@ int main(void)
printf("SINETABLE("#size") = {\n"); \
write_int32_t_array(sine_ ## size ## _fixed, size); \
printf("};\n")
+ PRINT_TABLE(120);
PRINT_TABLE(128);
+ PRINT_TABLE(480);
PRINT_TABLE(512);
+ PRINT_TABLE(960);
PRINT_TABLE(1024);
return 0;
}
diff --git a/libavcodec/sinewin_fixed_tablegen.h b/libavcodec/sinewin_fixed_tablegen.h
index c52c90bed7..056735704c 100644
--- a/libavcodec/sinewin_fixed_tablegen.h
+++ b/libavcodec/sinewin_fixed_tablegen.h
@@ -44,8 +44,11 @@
#include "libavutil/attributes.h"
#define SINETABLE_CONST
+SINETABLE( 120);
SINETABLE( 128);
+SINETABLE( 480);
SINETABLE( 512);
+SINETABLE( 960);
SINETABLE(1024);
#define SIN_FIX(a) (int)floor((a) * 0x80000000 + 0.5)
@@ -59,8 +62,11 @@ static av_cold void sine_window_init_fixed(int *window, int n)
static av_cold void init_sine_windows_fixed(void)
{
+ sine_window_init_fixed(sine_120_fixed, 120);
sine_window_init_fixed(sine_128_fixed, 128);
+ sine_window_init_fixed(sine_480_fixed, 480);
sine_window_init_fixed(sine_512_fixed, 512);
+ sine_window_init_fixed(sine_960_fixed, 960);
sine_window_init_fixed(sine_1024_fixed, 1024);
}
#endif /* CONFIG_HARDCODED_TABLES */
--
2.37.2.609.g9ff673ca1a
[-- Attachment #3: Type: text/plain, Size: 251 bytes --]
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [FFmpeg-devel] [PATCH] aacdec: convert to lavu/tx and support fixed-point 960-sample decoding
2022-10-28 21:46 [FFmpeg-devel] [PATCH] aacdec: convert to lavu/tx and support fixed-point 960-sample decoding Lynne
@ 2022-10-30 19:39 ` Michael Niedermayer
0 siblings, 0 replies; 2+ messages in thread
From: Michael Niedermayer @ 2022-10-30 19:39 UTC (permalink / raw)
To: FFmpeg development discussions and patches
[-- Attachment #1.1: Type: text/plain, Size: 2550 bytes --]
On Fri, Oct 28, 2022 at 11:46:58PM +0200, Lynne wrote:
> This patch replaces the transform used in AAC with lavu/tx and removes
> the limitation on only being able to decode 960-sample files
> with the float decoder.
> This commit also removes a whole bunch of unnecessary and slow
> lifting steps the decoder did to compensate for the poor accuracy
> of the old integer transformation code.
>
> Overall float decoder speedup on Zen 3 for 64kbps: 32%
>
> Patch attached.
>
> Note: lavu/tx's PFA 960pt transform currently has a bug that'll get fixed
> by my next patchset. The naive transforms are unaffected and may
> be used to test 960pt fixed-point decoding.
> configure | 4 -
> libavcodec/aac.h | 27 +++++----
> libavcodec/aac_defines.h | 4 -
> libavcodec/aacdec.c | 4 -
> libavcodec/aacdec_fixed.c | 5 +
> libavcodec/aacdec_template.c | 106 +++++++++++++-----------------------
> libavcodec/aacsbr.c | 1
> libavcodec/aacsbr_fixed.c | 1
> libavcodec/mips/aacdec_mips.c | 4 -
> libavcodec/sinewin_fixed_tablegen.c | 3 +
> libavcodec/sinewin_fixed_tablegen.h | 6 ++
> 11 files changed, 74 insertions(+), 91 deletions(-)
> 941b6d4662dd9808266b80fbafb5f003a734a415 0001-aacdec-convert-to-lavu-tx-and-support-fixed-point-96.patch
> From 95625b470fb61676e838310dc0108eb648f972a3 Mon Sep 17 00:00:00 2001
> From: Lynne <dev@lynne.ee>
> Date: Fri, 28 Oct 2022 21:36:10 +0200
> Subject: [PATCH] aacdec: convert to lavu/tx and support fixed-point 960-sample
> decoding
>
> This patch replaces the transform used in AAC with lavu/tx and removes
> the limitation on only being able to decode 960-sample files
> with the float decoder.
> This commit also removes a whole bunch of unnecessary and slow
> lifting steps the decoder did to compensate for the poor accuracy
> of the old integer transformation code.
>
> Overall float decoder speedup on Zen 3 for 64kbps: 32%
This needs a update to fate checksums here:
make: *** [fate-filter-meta-4560-rotate0] Error 1
make: *** [fate-prores-gray] Error 1
make: *** [fate-autorotate] Error 1
thx
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
Many that live deserve death. And some that die deserve life. Can you give
it to them? Then do not be too eager to deal out death in judgement. For
even the very wise cannot see all ends. -- Gandalf
[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 195 bytes --]
[-- Attachment #2: Type: text/plain, Size: 251 bytes --]
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-10-30 19:39 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-10-28 21:46 [FFmpeg-devel] [PATCH] aacdec: convert to lavu/tx and support fixed-point 960-sample decoding Lynne
2022-10-30 19:39 ` Michael Niedermayer
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git