mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2026-04-25 06:46:40 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

resample: use independent accumulators for SSE and SSSE3

Browse source
This commit is contained in:
Wim Taymans 2026-04-23 09:18:08 +02:00
parent 37f9f7773c
commit 75e432a49a
2 changed files with 22 additions and 41 deletions
Download patch file Download diff file Expand all files Collapse all files

38
spa/plugins/audioconvert/resample-native-sse.c
View file

@ -9,43 +9,23 @@
static inline void inner_product_sse(float *d, const float * SPA_RESTRICT s, static inline void inner_product_sse(float *d, const float * SPA_RESTRICT s,
const float * SPA_RESTRICT taps, uint32_t n_taps) const float * SPA_RESTRICT taps, uint32_t n_taps)
{ {
__m128 sum = _mm_setzero_ps(); __m128 sum[2] = { _mm_setzero_ps(), _mm_setzero_ps() };
uint32_t i = 0; uint32_t i;
#if 0
uint32_t unrolled = n_taps & ~15;
for (i = 0; i < unrolled; i += 16) { for (i = 0; i < n_taps; i += 8) {
sum = _mm_add_ps(sum, sum[0] = _mm_add_ps(sum[0],
_mm_mul_ps( _mm_mul_ps(
_mm_loadu_ps(s + i + 0), _mm_loadu_ps(s + i + 0),
_mm_load_ps(taps + i + 0))); _mm_load_ps(taps + i + 0)));
sum = _mm_add_ps(sum, sum[1] = _mm_add_ps(sum[1],
_mm_mul_ps(
_mm_loadu_ps(s + i + 4),
_mm_load_ps(taps + i + 4)));
sum = _mm_add_ps(sum,
_mm_mul_ps(
_mm_loadu_ps(s + i + 8),
_mm_load_ps(taps + i + 8)));
sum = _mm_add_ps(sum,
_mm_mul_ps(
_mm_loadu_ps(s + i + 12),
_mm_load_ps(taps + i + 12)));
}
#endif
for (; i < n_taps; i += 8) {
sum = _mm_add_ps(sum,
_mm_mul_ps(
_mm_loadu_ps(s + i + 0),
_mm_load_ps(taps + i + 0)));
sum = _mm_add_ps(sum,
_mm_mul_ps( _mm_mul_ps(
_mm_loadu_ps(s + i + 4), _mm_loadu_ps(s + i + 4),
_mm_load_ps(taps + i + 4))); _mm_load_ps(taps + i + 4)));
} }
sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); sum[0] = _mm_add_ps(sum[0], sum[1]);
sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); sum[0] = _mm_add_ps(sum[0], _mm_movehl_ps(sum[0], sum[0]));
_mm_store_ss(d, sum); sum[0] = _mm_add_ss(sum[0], _mm_shuffle_ps(sum[0], sum[0], 0x55));
_mm_store_ss(d, sum[0]);
} }
static inline void inner_product_ip_sse(float *d, const float * SPA_RESTRICT s, static inline void inner_product_ip_sse(float *d, const float * SPA_RESTRICT s,

25
spa/plugins/audioconvert/resample-native-ssse3.c
View file

@ -9,18 +9,18 @@
static inline void inner_product_ssse3(float *d, const float * SPA_RESTRICT s, static inline void inner_product_ssse3(float *d, const float * SPA_RESTRICT s,
const float * SPA_RESTRICT taps, uint32_t n_taps) const float * SPA_RESTRICT taps, uint32_t n_taps)
{ {
__m128 sum = _mm_setzero_ps(); __m128 sum[2] = { _mm_setzero_ps(), _mm_setzero_ps() };
__m128 t0, t1; __m128 t0, t1;
uint32_t i; uint32_t i;
switch (SPA_PTR_ALIGNMENT(s, 16)) { switch (SPA_PTR_ALIGNMENT(s, 16)) {
case 0: case 0:
for (i = 0; i < n_taps; i += 8) { for (i = 0; i < n_taps; i += 8) {
sum = _mm_add_ps(sum, sum[0] = _mm_add_ps(sum[0],
_mm_mul_ps( _mm_mul_ps(
_mm_load_ps(s + i + 0), _mm_load_ps(s + i + 0),
_mm_load_ps(taps + i + 0))); _mm_load_ps(taps + i + 0)));
sum = _mm_add_ps(sum, sum[1] = _mm_add_ps(sum[1],
_mm_mul_ps( _mm_mul_ps(
_mm_load_ps(s + i + 4), _mm_load_ps(s + i + 4),
_mm_load_ps(taps + i + 4))); _mm_load_ps(taps + i + 4)));
@ -31,12 +31,12 @@ static inline void inner_product_ssse3(float *d, const float * SPA_RESTRICT s,
for (i = 0; i < n_taps; i += 8) { for (i = 0; i < n_taps; i += 8) {
t1 = _mm_load_ps(s + i + 3); t1 = _mm_load_ps(s + i + 3);
t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 4); t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 4);
sum = _mm_add_ps(sum, sum[0] = _mm_add_ps(sum[0],
_mm_mul_ps(t0, _mm_load_ps(taps + i + 0))); _mm_mul_ps(t0, _mm_load_ps(taps + i + 0)));
t0 = t1; t0 = t1;
t1 = _mm_load_ps(s + i + 7); t1 = _mm_load_ps(s + i + 7);
t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 4); t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 4);
sum = _mm_add_ps(sum, sum[1] = _mm_add_ps(sum[1],
_mm_mul_ps(t0, _mm_load_ps(taps + i + 4))); _mm_mul_ps(t0, _mm_load_ps(taps + i + 4)));
t0 = t1; t0 = t1;
} }
@ -46,12 +46,12 @@ static inline void inner_product_ssse3(float *d, const float * SPA_RESTRICT s,
for (i = 0; i < n_taps; i += 8) { for (i = 0; i < n_taps; i += 8) {
t1 = _mm_load_ps(s + i + 2); t1 = _mm_load_ps(s + i + 2);
t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 8); t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 8);
sum = _mm_add_ps(sum, sum[0] = _mm_add_ps(sum[0],
_mm_mul_ps(t0, _mm_load_ps(taps + i + 0))); _mm_mul_ps(t0, _mm_load_ps(taps + i + 0)));
t0 = t1; t0 = t1;
t1 = _mm_load_ps(s + i + 6); t1 = _mm_load_ps(s + i + 6);
t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 8); t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 8);
sum = _mm_add_ps(sum, sum[1] = _mm_add_ps(sum[1],
_mm_mul_ps(t0, _mm_load_ps(taps + i + 4))); _mm_mul_ps(t0, _mm_load_ps(taps + i + 4)));
t0 = t1; t0 = t1;
} }
@ -61,20 +61,21 @@ static inline void inner_product_ssse3(float *d, const float * SPA_RESTRICT s,
for (i = 0; i < n_taps; i += 8) { for (i = 0; i < n_taps; i += 8) {
t1 = _mm_load_ps(s + i + 1); t1 = _mm_load_ps(s + i + 1);
t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 12); t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 12);
sum = _mm_add_ps(sum, sum[0] = _mm_add_ps(sum[0],
_mm_mul_ps(t0, _mm_load_ps(taps + i + 0))); _mm_mul_ps(t0, _mm_load_ps(taps + i + 0)));
t0 = t1; t0 = t1;
t1 = _mm_load_ps(s + i + 5); t1 = _mm_load_ps(s + i + 5);
t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 12); t0 = (__m128)_mm_alignr_epi8((__m128i)t1, (__m128i)t0, 12);
sum = _mm_add_ps(sum, sum[1] = _mm_add_ps(sum[1],
_mm_mul_ps(t0, _mm_load_ps(taps + i + 4))); _mm_mul_ps(t0, _mm_load_ps(taps + i + 4)));
t0 = t1; t0 = t1;
} }
break; break;
} }
sum = _mm_add_ps(sum, _mm_movehdup_ps(sum)); sum[0] = _mm_add_ps(sum[0], sum[1]);
sum = _mm_add_ss(sum, _mm_movehl_ps(sum, sum)); sum[0] = _mm_add_ps(sum[0], _mm_movehdup_ps(sum[0]));
_mm_store_ss(d, sum); sum[0] = _mm_add_ss(sum[0], _mm_movehl_ps(sum[0], sum[0]));
_mm_store_ss(d, sum[0]);
} }
static inline void inner_product_ip_ssse3(float *d, const float * SPA_RESTRICT s, static inline void inner_product_ip_ssse3(float *d, const float * SPA_RESTRICT s,