Implement a 6-point sinc-lanczos filter

7 files changed, 42 insertions, 21 deletions

diff --git a/Alc/ALc.c b/Alc/ALc.c
index fcb6504b..b9c01128 100644
--- a/Alc/ALc.c
+++ b/Alc/ALc.c
@@ -1011,9 +1011,11 @@ static void alc_initconfig(void)
             DefaultResampler = LinearResampler;
         else if(strcasecmp(str, "sinc4") == 0)
             DefaultResampler = FIR4Resampler;
+        else if(strcasecmp(str, "sinc6") == 0)
+            DefaultResampler = FIR6Resampler;
         else if(strcasecmp(str, "cubic") == 0)
         {
-            ERR("Resampler option \"cubic\" is deprecated, using sinc4\n");
+            WARN("Resampler option \"cubic\" is deprecated, using sinc4\n");
             DefaultResampler = FIR4Resampler;
         }
         else
diff --git a/Alc/ALu.c b/Alc/ALu.c
index 3ffe2e63..d8d41e2b 100644
--- a/Alc/ALu.c
+++ b/Alc/ALu.c
@@ -80,6 +80,7 @@ extern inline ALuint64 clampu64(ALuint64 val, ALuint64 min, ALuint64 max);
 
 extern inline ALfloat lerp(ALfloat val1, ALfloat val2, ALfloat mu);
 extern inline ALfloat resample_fir4(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALuint frac);
+extern inline ALfloat resample_fir6(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALfloat val4, ALfloat val5, ALuint frac);
 
 extern inline void aluVectorSet(aluVector *restrict vector, ALfloat x, ALfloat y, ALfloat z, ALfloat w);
 
diff --git a/Alc/mixer.c b/Alc/mixer.c
index b2bd1dce..dedcb5b9 100644
--- a/Alc/mixer.c
+++ b/Alc/mixer.c
@@ -43,7 +43,7 @@ static_assert((INT_MAX>>FRACTIONBITS)/MAX_PITCH > BUFFERSIZE,
 
 extern inline void InitiatePositionArrays(ALuint frac, ALuint increment, ALuint *frac_arr, ALuint *pos_arr, ALuint size);
 
-alignas(16) ALfloat ResampleCoeffs[FRACTIONONE][4];
+alignas(16) union ResamplerCoeffs ResampleCoeffs;
 
 
 static HrtfMixerFunc MixHrtfSamples = MixHrtf_C;
@@ -104,6 +104,8 @@ static inline ResamplerFunc SelectResampler(enum Resampler resampler)
                 return Resample_fir4_32_SSE2;
 #endif
             return Resample_fir4_32_C;
+        case FIR6Resampler:
+            return Resample_fir6_32_C;
         case ResamplerMax:
             /* Shouldn't happen */
             break;
@@ -113,27 +115,39 @@ static inline ResamplerFunc SelectResampler(enum Resampler resampler)
 }
 
 
-static float lanc2(float x)
+static float lanc(float r, float x)
 {
     if(x == 0.0f)
         return 1.0f;
-    if(fabsf(x) >= 2.0f)
+    if(fabsf(x) >= r)
         return 0.0f;
-    return 2.0f*sinf(x*F_PI)*sinf(x*F_PI/2.0f) /
+    return r*sinf(x*F_PI)*sinf(x*F_PI/r) /
            (F_PI*F_PI * x*x);
 }
 
 void aluInitMixer(void)
 {
     ALuint i;
-    for(i = 0;i < FRACTIONONE;i++)
-    {
-        ALfloat mu = (ALfloat)i / FRACTIONONE;
-        ResampleCoeffs[i][0] = lanc2(mu - -1.0f);
-        ResampleCoeffs[i][1] = lanc2(mu -  0.0f);
-        ResampleCoeffs[i][2] = lanc2(mu -  1.0f);
-        ResampleCoeffs[i][3] = lanc2(mu -  2.0f);
-    }
+    if(DefaultResampler == FIR6Resampler)
+        for(i = 0;i < FRACTIONONE;i++)
+        {
+            ALfloat mu = (ALfloat)i / FRACTIONONE;
+            ResampleCoeffs.FIR6[i][0] = lanc(3.0f, mu - -2.0f);
+            ResampleCoeffs.FIR6[i][1] = lanc(3.0f, mu - -1.0f);
+            ResampleCoeffs.FIR6[i][2] = lanc(3.0f, mu -  0.0f);
+            ResampleCoeffs.FIR6[i][3] = lanc(3.0f, mu -  1.0f);
+            ResampleCoeffs.FIR6[i][4] = lanc(3.0f, mu -  2.0f);
+            ResampleCoeffs.FIR6[i][5] = lanc(3.0f, mu -  3.0f);
+        }
+    else if(DefaultResampler == FIR4Resampler)
+        for(i = 0;i < FRACTIONONE;i++)
+        {
+            ALfloat mu = (ALfloat)i / FRACTIONONE;
+            ResampleCoeffs.FIR4[i][0] = lanc(2.0f, mu - -1.0f);
+            ResampleCoeffs.FIR4[i][1] = lanc(2.0f, mu -  0.0f);
+            ResampleCoeffs.FIR4[i][2] = lanc(2.0f, mu -  1.0f);
+            ResampleCoeffs.FIR4[i][3] = lanc(2.0f, mu -  2.0f);
+        }
 
     MixHrtfSamples = SelectHrtfMixer();
     MixSamples = SelectMixer();
diff --git a/Alc/mixer_c.c b/Alc/mixer_c.c
index 6164eb00..59ee5ebf 100644
--- a/Alc/mixer_c.c
+++ b/Alc/mixer_c.c
@@ -14,6 +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); }
 
 const ALfloat *Resample_copy32_C(const ALfloat *src, ALuint UNUSED(frac),
   ALuint UNUSED(increment), ALfloat *restrict dst, ALuint numsamples)
@@ -46,6 +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)
 
 #undef DECL_TEMPLATE
 
diff --git a/Alc/mixer_defs.h b/Alc/mixer_defs.h
index e577fafc..4c94954f 100644
--- a/Alc/mixer_defs.h
+++ b/Alc/mixer_defs.h
@@ -16,6 +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);
 
 
 /* C mixers */
diff --git a/Alc/mixer_sse2.c b/Alc/mixer_sse2.c
index 09a797c3..a6f2bb4e 100644
--- a/Alc/mixer_sse2.c
+++ b/Alc/mixer_sse2.c
@@ -100,10 +100,10 @@ const ALfloat *Resample_fir4_32_SSE2(const ALfloat *src, ALuint frac, ALuint inc
         const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]]);
         const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]]);
         const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]]);
-        __m128 k0 = _mm_load_ps(ResampleCoeffs[frac_.i[0]]);
-        __m128 k1 = _mm_load_ps(ResampleCoeffs[frac_.i[1]]);
-        __m128 k2 = _mm_load_ps(ResampleCoeffs[frac_.i[2]]);
-        __m128 k3 = _mm_load_ps(ResampleCoeffs[frac_.i[3]]);
+        __m128 k0 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[0]]);
+        __m128 k1 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[1]]);
+        __m128 k2 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[2]]);
+        __m128 k3 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[3]]);
         __m128 out;
 
         k0 = _mm_mul_ps(k0, val0);
diff --git a/Alc/mixer_sse41.c b/Alc/mixer_sse41.c
index c85a733d..b9ef608a 100644
--- a/Alc/mixer_sse41.c
+++ b/Alc/mixer_sse41.c
@@ -104,10 +104,10 @@ const ALfloat *Resample_fir4_32_SSE41(const ALfloat *src, ALuint frac, ALuint in
         const __m128 val1 = _mm_loadu_ps(&src[pos_.i[1]]);
         const __m128 val2 = _mm_loadu_ps(&src[pos_.i[2]]);
         const __m128 val3 = _mm_loadu_ps(&src[pos_.i[3]]);
-        __m128 k0 = _mm_load_ps(ResampleCoeffs[frac_.i[0]]);
-        __m128 k1 = _mm_load_ps(ResampleCoeffs[frac_.i[1]]);
-        __m128 k2 = _mm_load_ps(ResampleCoeffs[frac_.i[2]]);
-        __m128 k3 = _mm_load_ps(ResampleCoeffs[frac_.i[3]]);
+        __m128 k0 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[0]]);
+        __m128 k1 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[1]]);
+        __m128 k2 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[2]]);
+        __m128 k3 = _mm_load_ps(ResampleCoeffs.FIR4[frac_.i[3]]);
         __m128 out;
 
         k0 = _mm_mul_ps(k0, val0);