2017-06-18 13:06:30 +02:00
|
|
|
/*
|
|
|
|
* SIMD optimized non-power-of-two MDCT functions
|
|
|
|
*
|
|
|
|
* Copyright (C) 2017 Rostislav Pehlivanov <atomnuker@gmail.com>
|
|
|
|
*
|
|
|
|
* 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
|
|
|
|
*/
|
|
|
|
|
2021-07-23 17:50:51 +02:00
|
|
|
#include <string.h>
|
|
|
|
|
2017-06-18 13:06:30 +02:00
|
|
|
#include "config.h"
|
|
|
|
|
2021-04-14 15:11:43 +02:00
|
|
|
#include "libavutil/attributes.h"
|
2017-06-18 13:06:30 +02:00
|
|
|
#include "libavutil/x86/cpu.h"
|
|
|
|
#include "libavcodec/mdct15.h"
|
|
|
|
|
mdct15: add inverse transform postrotation SIMD
2.5ms frames:
Before (c): 2638 decicycles in postrotate, 2097040 runs, 112 skips
After (sse3): 1467 decicycles in postrotate, 2097083 runs, 69 skips
After (avx2): 1244 decicycles in postrotate, 2097085 runs, 67 skips
5ms frames:
Before (c): 4987 decicycles in postrotate, 1048371 runs, 205 skips
After (sse3): 2644 decicycles in postrotate, 1048509 runs, 67 skips
After (avx2): 2031 decicycles in postrotate, 1048523 runs, 53 skips
10ms frames:
Before (c): 9153 decicycles in postrotate, 523575 runs, 713 skips
After (sse3): 5110 decicycles in postrotate, 523726 runs, 562 skips
After (avx2): 3738 decicycles in postrotate, 524223 runs, 65 skips
20ms frames:
Before (c): 17857 decicycles in postrotate, 261866 runs, 278 skips
After (sse3): 10041 decicycles in postrotate, 261746 runs, 398 skips
After (avx2): 7050 decicycles in postrotate, 262116 runs, 28 skips
Improves total decoding performance for real world content by 9% with avx2.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2017-07-29 22:27:01 +02:00
|
|
|
void ff_mdct15_postreindex_sse3(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8);
|
|
|
|
void ff_mdct15_postreindex_avx2(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8);
|
|
|
|
|
2017-06-18 13:06:30 +02:00
|
|
|
void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride);
|
|
|
|
|
|
|
|
static void perm_twiddles(MDCT15Context *s)
|
|
|
|
{
|
|
|
|
int k;
|
2018-05-08 00:16:37 +02:00
|
|
|
FFTComplex tmp[30];
|
2017-06-18 13:06:30 +02:00
|
|
|
|
2018-05-08 00:16:37 +02:00
|
|
|
/* 5-point FFT twiddles */
|
|
|
|
s->exptab[60].re = s->exptab[60].im = s->exptab[19].re;
|
|
|
|
s->exptab[61].re = s->exptab[61].im = s->exptab[19].im;
|
|
|
|
s->exptab[62].re = s->exptab[62].im = s->exptab[20].re;
|
|
|
|
s->exptab[63].re = s->exptab[63].im = s->exptab[20].im;
|
2017-06-18 13:06:30 +02:00
|
|
|
|
|
|
|
/* 15-point FFT twiddles */
|
|
|
|
for (k = 0; k < 5; k++) {
|
2018-05-08 00:16:37 +02:00
|
|
|
tmp[6*k + 0] = s->exptab[k + 0];
|
|
|
|
tmp[6*k + 2] = s->exptab[k + 5];
|
|
|
|
tmp[6*k + 4] = s->exptab[k + 10];
|
2017-06-18 13:06:30 +02:00
|
|
|
|
2018-05-08 00:16:37 +02:00
|
|
|
tmp[6*k + 1] = s->exptab[2 * (k + 0)];
|
|
|
|
tmp[6*k + 3] = s->exptab[2 * (k + 5)];
|
|
|
|
tmp[6*k + 5] = s->exptab[2 * k + 5 ];
|
2017-06-18 13:06:30 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
for (k = 0; k < 6; k++) {
|
|
|
|
FFTComplex ac_exp[] = {
|
2018-05-08 00:16:37 +02:00
|
|
|
{ tmp[6*1 + k].re, tmp[6*1 + k].re },
|
|
|
|
{ tmp[6*2 + k].re, tmp[6*2 + k].re },
|
|
|
|
{ tmp[6*3 + k].re, tmp[6*3 + k].re },
|
|
|
|
{ tmp[6*4 + k].re, tmp[6*4 + k].re },
|
|
|
|
{ tmp[6*1 + k].im, -tmp[6*1 + k].im },
|
|
|
|
{ tmp[6*2 + k].im, -tmp[6*2 + k].im },
|
|
|
|
{ tmp[6*3 + k].im, -tmp[6*3 + k].im },
|
|
|
|
{ tmp[6*4 + k].im, -tmp[6*4 + k].im },
|
2017-06-18 13:06:30 +02:00
|
|
|
};
|
|
|
|
memcpy(s->exptab + 8*k, ac_exp, 8*sizeof(FFTComplex));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Specialcase when k = 0 */
|
|
|
|
for (k = 0; k < 3; k++) {
|
|
|
|
FFTComplex dc_exp[] = {
|
2018-05-08 00:16:37 +02:00
|
|
|
{ tmp[2*k + 0].re, -tmp[2*k + 0].im },
|
|
|
|
{ tmp[2*k + 0].im, tmp[2*k + 0].re },
|
|
|
|
{ tmp[2*k + 1].re, -tmp[2*k + 1].im },
|
|
|
|
{ tmp[2*k + 1].im, tmp[2*k + 1].re },
|
2017-06-18 13:06:30 +02:00
|
|
|
};
|
|
|
|
memcpy(s->exptab + 8*6 + 4*k, dc_exp, 4*sizeof(FFTComplex));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
av_cold void ff_mdct15_init_x86(MDCT15Context *s)
|
|
|
|
{
|
|
|
|
int adjust_twiddles = 0;
|
|
|
|
int cpu_flags = av_get_cpu_flags();
|
|
|
|
|
mdct15: add inverse transform postrotation SIMD
2.5ms frames:
Before (c): 2638 decicycles in postrotate, 2097040 runs, 112 skips
After (sse3): 1467 decicycles in postrotate, 2097083 runs, 69 skips
After (avx2): 1244 decicycles in postrotate, 2097085 runs, 67 skips
5ms frames:
Before (c): 4987 decicycles in postrotate, 1048371 runs, 205 skips
After (sse3): 2644 decicycles in postrotate, 1048509 runs, 67 skips
After (avx2): 2031 decicycles in postrotate, 1048523 runs, 53 skips
10ms frames:
Before (c): 9153 decicycles in postrotate, 523575 runs, 713 skips
After (sse3): 5110 decicycles in postrotate, 523726 runs, 562 skips
After (avx2): 3738 decicycles in postrotate, 524223 runs, 65 skips
20ms frames:
Before (c): 17857 decicycles in postrotate, 261866 runs, 278 skips
After (sse3): 10041 decicycles in postrotate, 261746 runs, 398 skips
After (avx2): 7050 decicycles in postrotate, 262116 runs, 28 skips
Improves total decoding performance for real world content by 9% with avx2.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2017-07-29 22:27:01 +02:00
|
|
|
if (EXTERNAL_SSE3(cpu_flags))
|
|
|
|
s->postreindex = ff_mdct15_postreindex_sse3;
|
|
|
|
|
2022-06-12 05:51:12 +02:00
|
|
|
#if ARCH_X86_64
|
|
|
|
if (EXTERNAL_AVX(cpu_flags)) {
|
2017-06-18 13:06:30 +02:00
|
|
|
s->fft15 = ff_fft15_avx;
|
|
|
|
adjust_twiddles = 1;
|
|
|
|
}
|
|
|
|
|
2022-06-12 05:51:12 +02:00
|
|
|
if (EXTERNAL_AVX2_FAST(cpu_flags))
|
mdct15: add inverse transform postrotation SIMD
2.5ms frames:
Before (c): 2638 decicycles in postrotate, 2097040 runs, 112 skips
After (sse3): 1467 decicycles in postrotate, 2097083 runs, 69 skips
After (avx2): 1244 decicycles in postrotate, 2097085 runs, 67 skips
5ms frames:
Before (c): 4987 decicycles in postrotate, 1048371 runs, 205 skips
After (sse3): 2644 decicycles in postrotate, 1048509 runs, 67 skips
After (avx2): 2031 decicycles in postrotate, 1048523 runs, 53 skips
10ms frames:
Before (c): 9153 decicycles in postrotate, 523575 runs, 713 skips
After (sse3): 5110 decicycles in postrotate, 523726 runs, 562 skips
After (avx2): 3738 decicycles in postrotate, 524223 runs, 65 skips
20ms frames:
Before (c): 17857 decicycles in postrotate, 261866 runs, 278 skips
After (sse3): 10041 decicycles in postrotate, 261746 runs, 398 skips
After (avx2): 7050 decicycles in postrotate, 262116 runs, 28 skips
Improves total decoding performance for real world content by 9% with avx2.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2017-07-29 22:27:01 +02:00
|
|
|
s->postreindex = ff_mdct15_postreindex_avx2;
|
2022-06-12 05:51:12 +02:00
|
|
|
#endif
|
mdct15: add inverse transform postrotation SIMD
2.5ms frames:
Before (c): 2638 decicycles in postrotate, 2097040 runs, 112 skips
After (sse3): 1467 decicycles in postrotate, 2097083 runs, 69 skips
After (avx2): 1244 decicycles in postrotate, 2097085 runs, 67 skips
5ms frames:
Before (c): 4987 decicycles in postrotate, 1048371 runs, 205 skips
After (sse3): 2644 decicycles in postrotate, 1048509 runs, 67 skips
After (avx2): 2031 decicycles in postrotate, 1048523 runs, 53 skips
10ms frames:
Before (c): 9153 decicycles in postrotate, 523575 runs, 713 skips
After (sse3): 5110 decicycles in postrotate, 523726 runs, 562 skips
After (avx2): 3738 decicycles in postrotate, 524223 runs, 65 skips
20ms frames:
Before (c): 17857 decicycles in postrotate, 261866 runs, 278 skips
After (sse3): 10041 decicycles in postrotate, 261746 runs, 398 skips
After (avx2): 7050 decicycles in postrotate, 262116 runs, 28 skips
Improves total decoding performance for real world content by 9% with avx2.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2017-07-29 22:27:01 +02:00
|
|
|
|
2017-06-18 13:06:30 +02:00
|
|
|
if (adjust_twiddles)
|
|
|
|
perm_twiddles(s);
|
|
|
|
}
|