mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2026-03-21 05:33:49 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

audioconver: add avx channelmix copy function

Browse source
This commit is contained in:
Wim Taymans 2026-03-20 18:23:43 +01:00
parent c02cdcb5ce
commit 9ba0c3cfd3
4 changed files with 82 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

63
spa/plugins/audioconvert/channelmix-ops-avx.c Normal file
View file

@ -0,0 +1,63 @@
/* Spa */
/* SPDX-FileCopyrightText: Copyright © 2018 Wim Taymans */
/* SPDX-License-Identifier: MIT */
#include "channelmix-ops.h"
#include <immintrin.h>
#include <float.h>
#include <math.h>
static inline void clear_avx(float *d, uint32_t n_samples)
{
memset(d, 0, n_samples * sizeof(float));
}
static inline void copy_avx(float *d, const float *s, uint32_t n_samples)
{
spa_memcpy(d, s, n_samples * sizeof(float));
}
static inline void vol_avx(float *d, const float *s, float vol, uint32_t n_samples)
{
uint32_t n, unrolled;
if (vol == 0.0f) {
clear_avx(d, n_samples);
} else if (vol == 1.0f) {
copy_avx(d, s, n_samples);
} else {
__m256 t[4];
const __m256 v = _mm256_set1_ps(vol);
if (SPA_IS_ALIGNED(d, 32) &&
SPA_IS_ALIGNED(s, 32))
unrolled = n_samples & ~31;
else
unrolled = 0;
for(n = 0; n < unrolled; n += 32) {
t[0] = _mm256_load_ps(&s[n]);
t[1] = _mm256_load_ps(&s[n+8]);
t[2] = _mm256_load_ps(&s[n+16]);
t[3] = _mm256_load_ps(&s[n+24]);
_mm256_store_ps(&d[n], _mm256_mul_ps(t[0], v));
_mm256_store_ps(&d[n+8], _mm256_mul_ps(t[1], v));
_mm256_store_ps(&d[n+16], _mm256_mul_ps(t[2], v));
_mm256_store_ps(&d[n+24], _mm256_mul_ps(t[3], v));
}
for(; n < n_samples; n++) {
__m128 v = _mm_set1_ps(vol);
_mm_store_ss(&d[n], _mm_mul_ss(_mm_load_ss(&s[n]), v));
}
}
}
void channelmix_copy_avx(struct channelmix *mix, void * SPA_RESTRICT dst[],
const void * SPA_RESTRICT src[], uint32_t n_samples)
{
uint32_t i, n_dst = mix->dst_chan;
float **d = (float **)dst;
const float **s = (const float **)src;
for (i = 0; i < n_dst; i++)
vol_avx(d[i], s[i], mix->matrix[i][i], n_samples);
}

5
spa/plugins/audioconvert/channelmix-ops.c
View file

@ -36,6 +36,11 @@ static const struct channelmix_info {
uint32_t cpu_flags;
} channelmix_table[] =
{
#if defined (HAVE_AVX)
MAKE(2, MASK_MONO, 2, MASK_MONO, channelmix_copy_avx, SPA_CPU_FLAG_AVX),
MAKE(2, MASK_STEREO, 2, MASK_STEREO, channelmix_copy_avx, SPA_CPU_FLAG_AVX),
MAKE(EQ, 0, EQ, 0, channelmix_copy_avx, SPA_CPU_FLAG_AVX),
#endif
#if defined (HAVE_SSE)
MAKE(2, MASK_MONO, 2, MASK_MONO, channelmix_copy_sse, SPA_CPU_FLAG_SSE),
MAKE(2, MASK_STEREO, 2, MASK_STEREO, channelmix_copy_sse, SPA_CPU_FLAG_SSE),

4
spa/plugins/audioconvert/channelmix-ops.h
View file

@ -140,4 +140,8 @@ DEFINE_FUNCTION(f32_5p1_4, sse);
DEFINE_FUNCTION(f32_7p1_4, sse);
#endif
#if defined (HAVE_AVX)
DEFINE_FUNCTION(copy, avx);
#endif
#undef DEFINE_FUNCTION

10
spa/plugins/audioconvert/meson.build
View file

@ -72,6 +72,16 @@ if have_sse41
simd_cargs += ['-DHAVE_SSE41']
simd_dependencies += audioconvert_sse41
endif
if have_avx
audioconvert_avx = static_library('audioconvert_avx',
['channelmix-ops-avx.c'],
c_args : [avx_args, '-O3', '-DHAVE_AVX', simd_cargs],
dependencies : [ spa_dep ],
install : false
)
simd_cargs += ['-DHAVE_AVX']
simd_dependencies += audioconvert_avx
endif
if have_avx2 and have_fma
audioconvert_avx2_fma = static_library('audioconvert_avx2_fma',
['resample-native-avx2.c'],