2012-05-21 19:58:41 +03:00
|
|
|
/*
|
2012-06-09 00:02:54 +03:00
|
|
|
* This file is part of FFmpeg.
|
2012-05-21 19:58:41 +03:00
|
|
|
*
|
2012-06-09 00:02:54 +03:00
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2012-05-21 19:58:41 +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.
|
|
|
|
*
|
2012-06-09 00:02:54 +03:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2012-05-21 19:58:41 +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
|
2012-06-09 00:02:54 +03:00
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
2012-05-21 19:58:41 +03:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
2013-04-30 19:31:25 +03:00
|
|
|
#include "libavutil/attributes.h"
|
2012-05-21 19:58:41 +03:00
|
|
|
#include "libavutil/cpu.h"
|
|
|
|
#include "libavutil/float_dsp.h"
|
2012-08-29 20:01:05 +03:00
|
|
|
#include "cpu.h"
|
2012-05-21 19:58:41 +03:00
|
|
|
|
2013-03-26 20:41:24 +03:00
|
|
|
void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
|
|
|
|
int len);
|
|
|
|
void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
|
|
|
|
int len);
|
|
|
|
|
2018-09-13 15:04:46 +02:00
|
|
|
void ff_vector_dmul_sse2(double *dst, const double *src0, const double *src1,
|
|
|
|
int len);
|
|
|
|
void ff_vector_dmul_avx(double *dst, const double *src0, const double *src1,
|
|
|
|
int len);
|
|
|
|
|
2013-03-26 20:41:24 +03:00
|
|
|
void ff_vector_fmac_scalar_sse(float *dst, const float *src, float mul,
|
2012-05-21 19:58:41 +03:00
|
|
|
int len);
|
2013-03-26 20:41:24 +03:00
|
|
|
void ff_vector_fmac_scalar_avx(float *dst, const float *src, float mul,
|
2012-05-21 19:58:41 +03:00
|
|
|
int len);
|
2014-03-10 23:09:20 +03:00
|
|
|
void ff_vector_fmac_scalar_fma3(float *dst, const float *src, float mul,
|
|
|
|
int len);
|
2012-05-21 19:58:41 +03:00
|
|
|
|
2013-03-26 20:41:24 +03:00
|
|
|
void ff_vector_fmul_scalar_sse(float *dst, const float *src, float mul,
|
|
|
|
int len);
|
2012-09-23 01:41:25 +03:00
|
|
|
|
2017-04-10 17:17:03 +02:00
|
|
|
void ff_vector_dmac_scalar_sse2(double *dst, const double *src, double mul,
|
|
|
|
int len);
|
|
|
|
void ff_vector_dmac_scalar_avx(double *dst, const double *src, double mul,
|
|
|
|
int len);
|
|
|
|
void ff_vector_dmac_scalar_fma3(double *dst, const double *src, double mul,
|
|
|
|
int len);
|
|
|
|
|
2013-03-26 20:41:24 +03:00
|
|
|
void ff_vector_dmul_scalar_sse2(double *dst, const double *src,
|
|
|
|
double mul, int len);
|
|
|
|
void ff_vector_dmul_scalar_avx(double *dst, const double *src,
|
|
|
|
double mul, int len);
|
2012-09-24 22:00:53 +03:00
|
|
|
|
2014-06-08 10:05:16 +03:00
|
|
|
void ff_vector_fmul_window_sse(float *dst, const float *src0,
|
|
|
|
const float *src1, const float *win, int len);
|
|
|
|
|
2013-01-20 09:26:58 +03:00
|
|
|
void ff_vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
|
|
|
|
const float *src2, int len);
|
|
|
|
void ff_vector_fmul_add_avx(float *dst, const float *src0, const float *src1,
|
|
|
|
const float *src2, int len);
|
2014-03-10 23:09:20 +03:00
|
|
|
void ff_vector_fmul_add_fma3(float *dst, const float *src0, const float *src1,
|
|
|
|
const float *src2, int len);
|
2013-01-20 09:26:58 +03:00
|
|
|
|
2013-01-21 00:20:30 +03:00
|
|
|
void ff_vector_fmul_reverse_sse(float *dst, const float *src0,
|
|
|
|
const float *src1, int len);
|
|
|
|
void ff_vector_fmul_reverse_avx(float *dst, const float *src0,
|
|
|
|
const float *src1, int len);
|
2017-04-12 02:29:09 +02:00
|
|
|
void ff_vector_fmul_reverse_avx2(float *dst, const float *src0,
|
|
|
|
const float *src1, int len);
|
2013-01-21 00:20:30 +03:00
|
|
|
|
2013-01-21 02:41:52 +03:00
|
|
|
float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
|
2021-01-20 17:58:31 +02:00
|
|
|
float ff_scalarproduct_float_fma3(const float *v1, const float *v2, int order);
|
2013-01-21 02:41:52 +03:00
|
|
|
|
2024-03-10 18:14:55 +02:00
|
|
|
void ff_butterflies_float_sse(float *restrict src0, float *restrict src1, int len);
|
2013-04-12 22:07:01 +03:00
|
|
|
|
2013-04-30 19:31:25 +03:00
|
|
|
av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
|
2012-05-21 19:58:41 +03:00
|
|
|
{
|
2013-07-17 21:19:24 +03:00
|
|
|
int cpu_flags = av_get_cpu_flags();
|
2012-05-21 19:58:41 +03:00
|
|
|
|
2013-07-17 21:19:24 +03:00
|
|
|
if (EXTERNAL_SSE(cpu_flags)) {
|
2012-05-21 19:58:41 +03:00
|
|
|
fdsp->vector_fmul = ff_vector_fmul_sse;
|
2012-06-09 06:20:59 +03:00
|
|
|
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_sse;
|
2012-09-23 01:41:25 +03:00
|
|
|
fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_sse;
|
2014-06-08 10:05:16 +03:00
|
|
|
fdsp->vector_fmul_window = ff_vector_fmul_window_sse;
|
2013-01-20 09:26:58 +03:00
|
|
|
fdsp->vector_fmul_add = ff_vector_fmul_add_sse;
|
2013-01-21 00:20:30 +03:00
|
|
|
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
|
2013-01-21 02:41:52 +03:00
|
|
|
fdsp->scalarproduct_float = ff_scalarproduct_float_sse;
|
2013-04-12 22:07:01 +03:00
|
|
|
fdsp->butterflies_float = ff_butterflies_float_sse;
|
2012-05-21 19:58:41 +03:00
|
|
|
}
|
2013-07-17 21:19:24 +03:00
|
|
|
if (EXTERNAL_SSE2(cpu_flags)) {
|
2018-09-13 15:04:46 +02:00
|
|
|
fdsp->vector_dmul = ff_vector_dmul_sse2;
|
2017-04-10 17:17:03 +02:00
|
|
|
fdsp->vector_dmac_scalar = ff_vector_dmac_scalar_sse2;
|
2012-09-24 22:00:53 +03:00
|
|
|
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2;
|
|
|
|
}
|
2015-05-31 19:20:29 +02:00
|
|
|
if (EXTERNAL_AVX_FAST(cpu_flags)) {
|
2012-05-21 19:58:41 +03:00
|
|
|
fdsp->vector_fmul = ff_vector_fmul_avx;
|
2018-09-13 15:04:46 +02:00
|
|
|
fdsp->vector_dmul = ff_vector_dmul_avx;
|
2012-06-09 06:20:59 +03:00
|
|
|
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx;
|
2012-09-24 22:00:53 +03:00
|
|
|
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx;
|
2017-04-10 17:17:03 +02:00
|
|
|
fdsp->vector_dmac_scalar = ff_vector_dmac_scalar_avx;
|
2013-01-20 09:26:58 +03:00
|
|
|
fdsp->vector_fmul_add = ff_vector_fmul_add_avx;
|
2013-01-21 00:20:30 +03:00
|
|
|
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
|
2012-05-21 19:58:41 +03:00
|
|
|
}
|
2017-04-12 02:29:09 +02:00
|
|
|
if (EXTERNAL_AVX2_FAST(cpu_flags)) {
|
|
|
|
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx2;
|
|
|
|
}
|
2016-02-07 05:52:06 +02:00
|
|
|
if (EXTERNAL_FMA3_FAST(cpu_flags)) {
|
2014-03-10 23:09:20 +03:00
|
|
|
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_fma3;
|
|
|
|
fdsp->vector_fmul_add = ff_vector_fmul_add_fma3;
|
2017-04-10 17:17:03 +02:00
|
|
|
fdsp->vector_dmac_scalar = ff_vector_dmac_scalar_fma3;
|
2021-01-20 17:58:31 +02:00
|
|
|
fdsp->scalarproduct_float = ff_scalarproduct_float_fma3;
|
2014-03-10 23:09:20 +03:00
|
|
|
}
|
2012-05-21 19:58:41 +03:00
|
|
|
}
|