mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2025-11-13 13:30:05 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

audiomixer: optimize avx mixer some more

Browse source 
Add avx mixer to test and benchmark
Rework and unroll the avx mixer some more.
The SSE one is 10 times faster than the C one, The AVX is 20 times
faster. The SSE2 function is 5 times faster than the C one.
This commit is contained in:
Wim Taymans 2022-07-10 23:13:24 +02:00
parent 23984f8790
commit 8fe83e5304
5 changed files with 77 additions and 58 deletions
Download patch file Download diff file Expand all files Collapse all files

19
spa/plugins/audiomixer/mix-ops-sse2.c
View file

@ -48,18 +48,17 @@ mix_f64_sse2(struct mix_ops *ops, void * SPA_RESTRICT dst, const void * SPA_REST
__m128d in[4];
const double **s = (const double **)src;
double *d = dst;
bool aligned = true;
if (SPA_UNLIKELY(!SPA_IS_ALIGNED(dst, 16)))
aligned = false;
else {
for (i = 0; i < n_src && aligned; i++) {
if (SPA_UNLIKELY(!SPA_IS_ALIGNED(src[i], 16)))
aligned = false;
if (SPA_LIKELY(SPA_IS_ALIGNED(dst, 16))) {
unrolled = n_samples & ~15;
for (i = 0; i < n_src; i++) {
if (SPA_UNLIKELY(!SPA_IS_ALIGNED(src[i], 16))) {
unrolled = 0;
break;
}
}
}
unrolled = aligned ? n_samples & ~7 : 0;
} else
unrolled = 0;
for (n = 0; n < unrolled; n += 8) {
in[0] = _mm_load_pd(&s[0][n+0]);