Replace the sinc6 resampler with sinc8, and make SSE versions

1 files changed, 72 insertions, 0 deletions

diff --git a/Alc/mixer_sse41.c b/Alc/mixer_sse41.c
index 90438e13..8fd5c4b5 100644
--- a/Alc/mixer_sse41.c
+++ b/Alc/mixer_sse41.c
@@ -147,3 +147,75 @@ const ALfloat *Resample_fir4_32_SSE41(const ALfloat *src, ALuint frac, ALuint in
     }
     return dst;
 }
+
+const ALfloat *Resample_fir8_32_SSE41(const ALfloat *src, ALuint frac, ALuint increment,
+                                      ALfloat *restrict dst, ALuint numsamples)
+{
+    const __m128i increment4 = _mm_set1_epi32(increment*4);
+    const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
+    alignas(16) union { ALuint i[4]; float f[4]; } pos_;
+    alignas(16) union { ALuint i[4]; float f[4]; } frac_;
+    __m128i frac4, pos4;
+    ALuint pos;
+    ALuint i, j;
+
+    InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
+
+    frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
+    pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
+
+    src -= 3;
+    for(i = 0;numsamples-i > 3;i += 4)
+    {
+        __m128 out[2];
+        for(j = 0;j < 8;j+=4)
+        {
+            const __m128 val0 = _mm_loadu_ps(&src[pos_.i[0]+j]);
+            const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]+j]);
+            const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]+j]);
+            const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]+j]);
+            __m128 k0 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[0]][j]);
+            __m128 k1 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[1]][j]);
+            __m128 k2 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[2]][j]);
+            __m128 k3 = _mm_load_ps(&ResampleCoeffs.FIR8[frac_.i[3]][j]);
+
+            k0 = _mm_mul_ps(k0, val0);
+            k1 = _mm_mul_ps(k1, val1);
+            k2 = _mm_mul_ps(k2, val2);
+            k3 = _mm_mul_ps(k3, val3);
+            k0 = _mm_hadd_ps(k0, k1);
+            k2 = _mm_hadd_ps(k2, k3);
+            out[j>>2] = _mm_hadd_ps(k0, k2);
+        }
+
+        out[0] = _mm_add_ps(out[0], out[1]);
+        _mm_store_ps(&dst[i], out[0]);
+
+        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);
+        frac_.i[0] = _mm_extract_epi32(frac4, 0);
+        frac_.i[1] = _mm_extract_epi32(frac4, 1);
+        frac_.i[2] = _mm_extract_epi32(frac4, 2);
+        frac_.i[3] = _mm_extract_epi32(frac4, 3);
+    }
+
+    pos = pos_.i[0];
+    frac = frac_.i[0];
+
+    for(;i < numsamples;i++)
+    {
+        dst[i] = resample_fir8(src[pos  ], src[pos+1], src[pos+2], src[pos+3],
+                               src[pos+4], src[pos+5], src[pos+6], src[pos+7], frac);
+
+        frac += increment;
+        pos  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
+    }
+    return dst;
+}