mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2025-11-03 09:01:54 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

audioconvert: unroll some loop

Browse source
This commit is contained in:
Wim Taymans 2019-04-10 12:53:25 +02:00
parent 40a5063b9f
commit ae3e5787af
1 changed files with 23 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

34
spa/plugins/audioconvert/fmt-ops-sse2.c
View file

@ -80,32 +80,44 @@ conv_s16_to_f32d_2_sse2(struct convert *conv, void * SPA_RESTRICT dst[], const v
float **d = (float **) dst; float **d = (float **) dst;
float *d0 = d[0], *d1 = d[1]; float *d0 = d[0], *d1 = d[1];
uint32_t n, unrolled; uint32_t n, unrolled;
__m128i in, t[2]; __m128i in[2], t[4];
__m128 out[2], factor = _mm_set1_ps(1.0f / S16_SCALE); __m128 out[4], factor = _mm_set1_ps(1.0f / S16_SCALE);
if (SPA_IS_ALIGNED(s, 16) && if (SPA_IS_ALIGNED(s, 16) &&
SPA_IS_ALIGNED(d0, 16) && SPA_IS_ALIGNED(d0, 16) &&
SPA_IS_ALIGNED(d1, 16)) SPA_IS_ALIGNED(d1, 16))
unrolled = n_samples & ~3; unrolled = n_samples & ~7;
else else
unrolled = 0; unrolled = 0;
for(n = 0; n < unrolled; n += 4) { for(n = 0; n < unrolled; n += 8) {
in = _mm_load_si128((__m128i*)s); in[0] = _mm_load_si128((__m128i*)(s + 0));
in[1] = _mm_load_si128((__m128i*)(s + 8));
t[0] = _mm_slli_epi32(in, 16); t[0] = _mm_slli_epi32(in[0], 16);
t[0] = _mm_srai_epi32(t[0], 16); t[0] = _mm_srai_epi32(t[0], 16);
t[1] = _mm_srai_epi32(in, 16);
out[0] = _mm_cvtepi32_ps(t[0]); out[0] = _mm_cvtepi32_ps(t[0]);
out[0] = _mm_mul_ps(out[0], factor); out[0] = _mm_mul_ps(out[0], factor);
t[1] = _mm_srai_epi32(in[0], 16);
out[1] = _mm_cvtepi32_ps(t[1]); out[1] = _mm_cvtepi32_ps(t[1]);
out[1] = _mm_mul_ps(out[1], factor); out[1] = _mm_mul_ps(out[1], factor);
_mm_store_ps(&d0[n], out[0]); t[2] = _mm_slli_epi32(in[1], 16);
_mm_store_ps(&d1[n], out[1]); t[2] = _mm_srai_epi32(t[2], 16);
out[2] = _mm_cvtepi32_ps(t[2]);
out[2] = _mm_mul_ps(out[2], factor);
s += 8; t[3] = _mm_srai_epi32(in[1], 16);
out[3] = _mm_cvtepi32_ps(t[3]);
out[3] = _mm_mul_ps(out[3], factor);
_mm_store_ps(&d0[n + 0], out[0]);
_mm_store_ps(&d1[n + 0], out[1]);
_mm_store_ps(&d0[n + 4], out[2]);
_mm_store_ps(&d1[n + 4], out[3]);
s += 16;
} }
for(; n < n_samples; n++) { for(; n < n_samples; n++) {
out[0] = _mm_cvtsi32_ss(out[0], s[0]); out[0] = _mm_cvtsi32_ss(out[0], s[0]);