Replace the sinc6 resampler with sinc8, and make SSE versions

8 files changed, 177 insertions, 23 deletions

diff --git a/Alc/mixer.c b/Alc/mixer.c
index 8061ab73..2f38ef3f 100644
--- a/Alc/mixer.c
+++ b/Alc/mixer.c
@@ -50,7 +50,7 @@ enum Resampler {
     PointResampler,
     LinearResampler,
     FIR4Resampler,
-    FIR6Resampler,
+    FIR8Resampler,
 
     ResamplerMax,
 };
@@ -65,7 +65,7 @@ static const ALsizei ResamplerPadding[ResamplerMax][2] = {
     {0, 0}, /* Point */
     {0, 1}, /* Linear */
     {1, 2}, /* FIR4 */
-    {2, 3}, /* FIR6 */
+    {3, 4}, /* FIR8 */
 };
 
 
@@ -127,8 +127,16 @@ static inline ResamplerFunc SelectResampler(enum Resampler resampler)
                 return Resample_fir4_32_SSE3;
 #endif
             return Resample_fir4_32_C;
-        case FIR6Resampler:
-            return Resample_fir6_32_C;
+        case FIR8Resampler:
+#ifdef HAVE_SSE4_1
+            if((CPUCapFlags&CPU_CAP_SSE4_1))
+                return Resample_fir8_32_SSE41;
+#endif
+#ifdef HAVE_SSE3
+            if((CPUCapFlags&CPU_CAP_SSE3))
+                return Resample_fir8_32_SSE3;
+#endif
+            return Resample_fir8_32_C;
         case ResamplerMax:
             /* Shouldn't happen */
             break;
@@ -161,8 +169,8 @@ void aluInitMixer(void)
             DefaultResampler = LinearResampler;
         else if(strcasecmp(str, "sinc4") == 0)
             DefaultResampler = FIR4Resampler;
-        else if(strcasecmp(str, "sinc6") == 0)
-            DefaultResampler = FIR6Resampler;
+        else if(strcasecmp(str, "sinc8") == 0)
+            DefaultResampler = FIR8Resampler;
         else if(strcasecmp(str, "cubic") == 0)
         {
             WARN("Resampler option \"cubic\" is deprecated, using sinc4\n");
@@ -179,16 +187,18 @@ void aluInitMixer(void)
         }
     }
 
-    if(DefaultResampler == FIR6Resampler)
+    if(DefaultResampler == FIR8Resampler)
         for(i = 0;i < FRACTIONONE;i++)
         {
             ALdouble mu = (ALdouble)i / FRACTIONONE;
-            ResampleCoeffs.FIR6[i][0] = lanc(3.0, mu - -2.0);
-            ResampleCoeffs.FIR6[i][1] = lanc(3.0, mu - -1.0);
-            ResampleCoeffs.FIR6[i][2] = lanc(3.0, mu -  0.0);
-            ResampleCoeffs.FIR6[i][3] = lanc(3.0, mu -  1.0);
-            ResampleCoeffs.FIR6[i][4] = lanc(3.0, mu -  2.0);
-            ResampleCoeffs.FIR6[i][5] = lanc(3.0, mu -  3.0);
+            ResampleCoeffs.FIR8[i][0] = lanc(4.0, mu - -3.0);
+            ResampleCoeffs.FIR8[i][1] = lanc(4.0, mu - -2.0);
+            ResampleCoeffs.FIR8[i][2] = lanc(4.0, mu - -1.0);
+            ResampleCoeffs.FIR8[i][3] = lanc(4.0, mu -  0.0);
+            ResampleCoeffs.FIR8[i][4] = lanc(4.0, mu -  1.0);
+            ResampleCoeffs.FIR8[i][5] = lanc(4.0, mu -  2.0);
+            ResampleCoeffs.FIR8[i][6] = lanc(4.0, mu -  3.0);
+            ResampleCoeffs.FIR8[i][7] = lanc(4.0, mu -  4.0);
         }
     else if(DefaultResampler == FIR4Resampler)
         for(i = 0;i < FRACTIONONE;i++)
diff --git a/Alc/mixer_c.c b/Alc/mixer_c.c
index 0d3e99a6..c9fa730a 100644
--- a/Alc/mixer_c.c
+++ b/Alc/mixer_c.c
@@ -14,8 +14,8 @@ static inline ALfloat lerp32(const ALfloat *vals, ALuint frac)
 { return lerp(vals[0], vals[1], frac * (1.0f/FRACTIONONE)); }
 static inline ALfloat fir4_32(const ALfloat *vals, ALuint frac)
 { return resample_fir4(vals[-1], vals[0], vals[1], vals[2], frac); }
-static inline ALfloat fir6_32(const ALfloat *vals, ALuint frac)
-{ return resample_fir6(vals[-2], vals[-1], vals[0], vals[1], vals[2], vals[3], frac); }
+static inline ALfloat fir8_32(const ALfloat *vals, ALuint frac)
+{ return resample_fir8(vals[-3], vals[-2], vals[-1], vals[0], vals[1], vals[2], vals[3], vals[4], frac); }
 
 const ALfloat *Resample_copy32_C(const ALfloat *src, ALuint UNUSED(frac),
   ALuint UNUSED(increment), ALfloat *restrict dst, ALuint numsamples)
@@ -48,7 +48,7 @@ const ALfloat *Resample_##Sampler##_C(const ALfloat *src, ALuint frac,        \
 DECL_TEMPLATE(point32)
 DECL_TEMPLATE(lerp32)
 DECL_TEMPLATE(fir4_32)
-DECL_TEMPLATE(fir6_32)
+DECL_TEMPLATE(fir8_32)
 
 #undef DECL_TEMPLATE
 
diff --git a/Alc/mixer_defs.h b/Alc/mixer_defs.h
index bba8e9a1..9f3b4a6f 100644
--- a/Alc/mixer_defs.h
+++ b/Alc/mixer_defs.h
@@ -16,7 +16,7 @@ const ALfloat *Resample_copy32_C(const ALfloat *src, ALuint frac, ALuint increme
 const ALfloat *Resample_point32_C(const ALfloat *src, ALuint frac, ALuint increment, ALfloat *restrict dst, ALuint dstlen);
 const ALfloat *Resample_lerp32_C(const ALfloat *src, ALuint frac, ALuint increment, ALfloat *restrict dst, ALuint dstlen);
 const ALfloat *Resample_fir4_32_C(const ALfloat *src, ALuint frac, ALuint increment, ALfloat *restrict dst, ALuint dstlen);
-const ALfloat *Resample_fir6_32_C(const ALfloat *src, ALuint frac, ALuint increment, ALfloat *restrict dst, ALuint dstlen);
+const ALfloat *Resample_fir8_32_C(const ALfloat *src, ALuint frac, ALuint increment, ALfloat *restrict dst, ALuint dstlen);
 
 
 /* C mixers */
@@ -60,6 +60,11 @@ const ALfloat *Resample_fir4_32_SSE3(const ALfloat *src, ALuint frac, ALuint inc
 const ALfloat *Resample_fir4_32_SSE41(const ALfloat *src, ALuint frac, ALuint increment,
                                       ALfloat *restrict dst, ALuint numsamples);
 
+const ALfloat *Resample_fir8_32_SSE3(const ALfloat *src, ALuint frac, ALuint increment,
+                                     ALfloat *restrict dst, ALuint numsamples);
+const ALfloat *Resample_fir8_32_SSE41(const ALfloat *src, ALuint frac, ALuint increment,
+                                      ALfloat *restrict dst, ALuint numsamples);
+
 /* Neon mixers */
 void MixHrtf_Neon(ALfloat (*restrict OutBuffer)[BUFFERSIZE], const ALfloat *data,
                   ALuint Counter, ALuint Offset, ALuint OutPos, const ALuint IrSize,
diff --git a/Alc/mixer_sse3.c b/Alc/mixer_sse3.c
index ced90593..dbf963ff 100644
--- a/Alc/mixer_sse3.c
+++ b/Alc/mixer_sse3.c
@@ -91,3 +91,69 @@ const ALfloat *Resample_fir4_32_SSE3(const ALfloat *src, ALuint frac, ALuint inc
     }
     return dst;
 }
+
+const ALfloat *Resample_fir8_32_SSE3(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);
+
+        _mm_store_ps(pos_.f, _mm_castsi128_ps(pos4));
+        _mm_store_ps(frac_.f, _mm_castsi128_ps(frac4));
+    }
+
+    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;
+}
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;
+}
diff --git a/OpenAL32/Include/alu.h b/OpenAL32/Include/alu.h
index c37ebb9f..051a0f0d 100644
--- a/OpenAL32/Include/alu.h
+++ b/OpenAL32/Include/alu.h
@@ -204,7 +204,7 @@ inline ALuint64 clampu64(ALuint64 val, ALuint64 min, ALuint64 max)
 
 union ResamplerCoeffs {
     ALfloat FIR4[FRACTIONONE][4];
-    ALfloat FIR6[FRACTIONONE][6];
+    ALfloat FIR8[FRACTIONONE][8];
 };
 extern alignas(16) union ResamplerCoeffs ResampleCoeffs;
 
@@ -218,10 +218,11 @@ inline ALfloat resample_fir4(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat v
     const ALfloat *k = ResampleCoeffs.FIR4[frac];
     return k[0]*val0 + k[1]*val1 + k[2]*val2 + k[3]*val3;
 }
-inline ALfloat resample_fir6(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALfloat val4, ALfloat val5, ALuint frac)
+inline ALfloat resample_fir8(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALfloat val4, ALfloat val5, ALfloat val6, ALfloat val7, ALuint frac)
 {
-    const ALfloat *k = ResampleCoeffs.FIR6[frac];
-    return k[0]*val0 + k[1]*val1 + k[2]*val2 + k[3]*val3 + k[4]*val4 + k[5]*val5;
+    const ALfloat *k = ResampleCoeffs.FIR8[frac];
+    return k[0]*val0 + k[1]*val1 + k[2]*val2 + k[3]*val3 +
+           k[4]*val4 + k[5]*val5 + k[6]*val6 + k[7]*val7;
 }
 
 
diff --git a/alsoftrc.sample b/alsoftrc.sample
index 4dc2cd06..faac3d8e 100644
--- a/alsoftrc.sample
+++ b/alsoftrc.sample
@@ -130,7 +130,7 @@
 #  point - nearest sample, no interpolation
 #  linear - extrapolates samples using a linear slope between samples
 #  sinc4 - extrapolates samples using a 4-point Sinc-Lanczos filter
-#  sinc6 - extrapolates samples using a 6-point Sinc-Lanczos filter
+#  sinc8 - extrapolates samples using an 8-point Sinc-Lanczos filter
 #  Specifying other values will result in using the default (linear).
 #resampler = linear
 
diff --git a/utils/alsoft-config/mainwindow.cpp b/utils/alsoft-config/mainwindow.cpp
index 73773836..49a6c474 100644
--- a/utils/alsoft-config/mainwindow.cpp
+++ b/utils/alsoft-config/mainwindow.cpp
@@ -64,7 +64,7 @@ static const struct {
     { "Point (low quality, fast)", "point" },
     { "Linear (basic quality, fast)", "linear" },
     { "Sinc-Lanczos 4pt (good quality)", "sinc4" },
-    { "Sinc-Lanczos 6pt (high quality, slow)", "sinc6" },
+    { "Sinc-Lanczos 8pt (high quality, slow)", "sinc8" },
 
     { "", "" }
 }, stereoModeList[] = {