diff options
author | Chris Robinson <[email protected]> | 2018-09-03 12:18:06 -0700 |
---|---|---|
committer | Chris Robinson <[email protected]> | 2018-09-03 12:42:31 -0700 |
commit | fce86815f494b981f3cbbba290f5399c9907dee7 (patch) | |
tree | 3a83696340c7abcfbec0c041a8ad899759b90b14 /Alc/mixer/mixer_sse41.c | |
parent | 3b4f28d173d21bad20f28ec41f594f88459596dd (diff) |
Extract SIMD values right before using them
Diffstat (limited to 'Alc/mixer/mixer_sse41.c')
-rw-r--r-- | Alc/mixer/mixer_sse41.c | 15 |
1 files changed, 7 insertions, 8 deletions
diff --git a/Alc/mixer/mixer_sse41.c b/Alc/mixer/mixer_sse41.c index 98a3ef74..396b7c7c 100644 --- a/Alc/mixer/mixer_sse41.c +++ b/Alc/mixer/mixer_sse41.c @@ -50,8 +50,12 @@ const ALfloat *Resample_lerp_SSE41(const InterpState* UNUSED(state), for(i = 0;numsamples-i > 3;i += 4) { - const __m128 val1 = _mm_setr_ps(src[pos_.i[0]], src[pos_.i[1]], src[pos_.i[2]], src[pos_.i[3]]); - const __m128 val2 = _mm_setr_ps(src[pos_.i[0]+1], src[pos_.i[1]+1], src[pos_.i[2]+1], src[pos_.i[3]+1]); + const int pos0 = _mm_extract_epi32(pos4, 0); + const int pos1 = _mm_extract_epi32(pos4, 1); + const int pos2 = _mm_extract_epi32(pos4, 2); + const int pos3 = _mm_extract_epi32(pos4, 3); + const __m128 val1 = _mm_setr_ps(src[pos0 ], src[pos1 ], src[pos2 ], src[pos3 ]); + const __m128 val2 = _mm_setr_ps(src[pos0+1], src[pos1+1], src[pos2+1], src[pos3+1]); /* val1 + (val2-val1)*mu */ const __m128 r0 = _mm_sub_ps(val2, val1); @@ -63,17 +67,12 @@ const ALfloat *Resample_lerp_SSE41(const InterpState* UNUSED(state), frac4 = _mm_add_epi32(frac4, increment4); pos4 = _mm_add_epi32(pos4, _mm_srli_epi32(frac4, FRACTIONBITS)); frac4 = _mm_and_si128(frac4, fracMask4); - - pos_.i[0] = _mm_extract_epi32(pos4, 0); - pos_.i[1] = _mm_extract_epi32(pos4, 1); - pos_.i[2] = _mm_extract_epi32(pos4, 2); - pos_.i[3] = _mm_extract_epi32(pos4, 3); } /* NOTE: These four elements represent the position *after* the last four * samples, so the lowest element is the next position to resample. */ - pos = pos_.i[0]; + pos = _mm_cvtsi128_si32(pos4); frac = _mm_cvtsi128_si32(frac4); for(;i < numsamples;i++) |