c8af852b |
/*
* Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
* |
d814a839 |
* This file is part of FFmpeg. |
c8af852b |
* |
d814a839 |
* FFmpeg is free software; you can redistribute it and/or |
c8af852b |
* 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.
* |
d814a839 |
* FFmpeg is distributed in the hope that it will be useful, |
c8af852b |
* 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 |
d814a839 |
* License along with FFmpeg; if not, write to the Free Software |
c8af852b |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "config.h"
#include "libavutil/cpu.h" |
e0c6cce4 |
#include "libavutil/x86/cpu.h" |
c8af852b |
#include "libavresample/audio_mix.h"
|
b6649ab5 |
void ff_mix_2_to_1_fltp_flt_sse(float **src, float **matrix, int len,
int out_ch, int in_ch);
void ff_mix_2_to_1_fltp_flt_avx(float **src, float **matrix, int len,
int out_ch, int in_ch); |
c8af852b |
|
b6649ab5 |
void ff_mix_2_to_1_s16p_flt_sse2(int16_t **src, float **matrix, int len,
int out_ch, int in_ch);
void ff_mix_2_to_1_s16p_flt_sse4(int16_t **src, float **matrix, int len,
int out_ch, int in_ch); |
c140fb2c |
|
b6649ab5 |
void ff_mix_2_to_1_s16p_q8_sse2(int16_t **src, int16_t **matrix,
int len, int out_ch, int in_ch); |
b75726cb |
|
b6649ab5 |
void ff_mix_1_to_2_fltp_flt_sse(float **src, float **matrix, int len,
int out_ch, int in_ch);
void ff_mix_1_to_2_fltp_flt_avx(float **src, float **matrix, int len,
int out_ch, int in_ch); |
29f7490c |
|
b6649ab5 |
void ff_mix_1_to_2_s16p_flt_sse2(int16_t **src, float **matrix, int len,
int out_ch, int in_ch);
void ff_mix_1_to_2_s16p_flt_sse4(int16_t **src, float **matrix, int len,
int out_ch, int in_ch);
void ff_mix_1_to_2_s16p_flt_avx (int16_t **src, float **matrix, int len,
int out_ch, int in_ch); |
f61ce90c |
|
b6649ab5 |
#define DEFINE_MIX_3_8_TO_1_2(chan) \
void ff_mix_ ## chan ## _to_1_fltp_flt_sse(float **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_fltp_flt_sse(float **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
\
void ff_mix_ ## chan ## _to_1_s16p_flt_sse2(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_s16p_flt_sse2(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
\
void ff_mix_ ## chan ## _to_1_s16p_flt_sse4(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_s16p_flt_sse4(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
\
void ff_mix_ ## chan ## _to_1_fltp_flt_avx(float **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_fltp_flt_avx(float **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
\
void ff_mix_ ## chan ## _to_1_s16p_flt_avx(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_s16p_flt_avx(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
\
void ff_mix_ ## chan ## _to_1_fltp_flt_fma4(float **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_fltp_flt_fma4(float **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
\
void ff_mix_ ## chan ## _to_1_s16p_flt_fma4(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); \
void ff_mix_ ## chan ## _to_2_s16p_flt_fma4(int16_t **src, \
float **matrix, int len, \
int out_ch, int in_ch); |
2f096bb1 |
DEFINE_MIX_3_8_TO_1_2(3)
DEFINE_MIX_3_8_TO_1_2(4)
DEFINE_MIX_3_8_TO_1_2(5)
DEFINE_MIX_3_8_TO_1_2(6)
DEFINE_MIX_3_8_TO_1_2(7)
DEFINE_MIX_3_8_TO_1_2(8)
#define SET_MIX_3_8_TO_1_2(chan) \ |
3ac7fa81 |
if (EXTERNAL_SSE(cpu_flags)) { \ |
2f096bb1 |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, 16, 4, "SSE", \
ff_mix_ ## chan ## _to_1_fltp_flt_sse); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, 16, 4, "SSE", \
ff_mix_## chan ##_to_2_fltp_flt_sse); \
} \ |
3ac7fa81 |
if (EXTERNAL_SSE2(cpu_flags)) { \ |
2f096bb1 |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, 16, 8, "SSE2", \
ff_mix_ ## chan ## _to_1_s16p_flt_sse2); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, 16, 8, "SSE2", \
ff_mix_ ## chan ## _to_2_s16p_flt_sse2); \
} \ |
3ac7fa81 |
if (EXTERNAL_SSE4(cpu_flags)) { \ |
2f096bb1 |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, 16, 8, "SSE4", \
ff_mix_ ## chan ## _to_1_s16p_flt_sse4); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, 16, 8, "SSE4", \
ff_mix_ ## chan ## _to_2_s16p_flt_sse4); \
} \ |
3ac7fa81 |
if (EXTERNAL_AVX(cpu_flags)) { \ |
2f096bb1 |
int ptr_align = 32; \
int smp_align = 8; \
if (ARCH_X86_32 || chan >= 6) { \
ptr_align = 16; \
smp_align = 4; \
} \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, ptr_align, smp_align, "AVX", \
ff_mix_ ## chan ## _to_1_fltp_flt_avx); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, ptr_align, smp_align, "AVX", \
ff_mix_ ## chan ## _to_2_fltp_flt_avx); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, 16, 8, "AVX", \
ff_mix_ ## chan ## _to_1_s16p_flt_avx); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, 16, 8, "AVX", \
ff_mix_ ## chan ## _to_2_s16p_flt_avx); \
} \ |
3ac7fa81 |
if (EXTERNAL_FMA4(cpu_flags)) { \ |
2f096bb1 |
int ptr_align = 32; \
int smp_align = 8; \
if (ARCH_X86_32 || chan >= 6) { \
ptr_align = 16; \
smp_align = 4; \
} \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, ptr_align, smp_align, "FMA4", \
ff_mix_ ## chan ## _to_1_fltp_flt_fma4); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, ptr_align, smp_align, "FMA4", \
ff_mix_ ## chan ## _to_2_fltp_flt_fma4); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 1, 16, 8, "FMA4", \
ff_mix_ ## chan ## _to_1_s16p_flt_fma4); \
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
chan, 2, 16, 8, "FMA4", \
ff_mix_ ## chan ## _to_2_s16p_flt_fma4); \
}
|
c8af852b |
av_cold void ff_audio_mix_init_x86(AudioMix *am)
{ |
3ac7fa81 |
int cpu_flags = av_get_cpu_flags(); |
c8af852b |
|
3ac7fa81 |
if (EXTERNAL_SSE(cpu_flags)) { |
c8af852b |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
2, 1, 16, 8, "SSE", ff_mix_2_to_1_fltp_flt_sse); |
29f7490c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
1, 2, 16, 4, "SSE", ff_mix_1_to_2_fltp_flt_sse); |
c8af852b |
} |
3ac7fa81 |
if (EXTERNAL_SSE2(cpu_flags)) { |
c140fb2c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
2, 1, 16, 8, "SSE2", ff_mix_2_to_1_s16p_flt_sse2); |
b75726cb |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_Q8,
2, 1, 16, 8, "SSE2", ff_mix_2_to_1_s16p_q8_sse2); |
f61ce90c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
1, 2, 16, 8, "SSE2", ff_mix_1_to_2_s16p_flt_sse2); |
c140fb2c |
} |
3ac7fa81 |
if (EXTERNAL_SSE4(cpu_flags)) { |
c140fb2c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
2, 1, 16, 8, "SSE4", ff_mix_2_to_1_s16p_flt_sse4); |
f61ce90c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
1, 2, 16, 8, "SSE4", ff_mix_1_to_2_s16p_flt_sse4); |
c140fb2c |
} |
d68c0538 |
if (EXTERNAL_AVX_FAST(cpu_flags)) { |
c8af852b |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
2, 1, 32, 16, "AVX", ff_mix_2_to_1_fltp_flt_avx); |
29f7490c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,
1, 2, 32, 8, "AVX", ff_mix_1_to_2_fltp_flt_avx); |
d68c0538 |
}
if (EXTERNAL_AVX(cpu_flags)) { |
f61ce90c |
ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
1, 2, 16, 8, "AVX", ff_mix_1_to_2_s16p_flt_avx); |
c8af852b |
} |
2f096bb1 |
SET_MIX_3_8_TO_1_2(3)
SET_MIX_3_8_TO_1_2(4)
SET_MIX_3_8_TO_1_2(5)
SET_MIX_3_8_TO_1_2(6)
SET_MIX_3_8_TO_1_2(7)
SET_MIX_3_8_TO_1_2(8) |
c8af852b |
} |