Use a lookup table to do cubic resampling

Alc/ALc.c

@@ -1006,6 +1006,7 @@ static void alc_initconfig(void)
                 WARN("Invalid resampler: %s\n", str);
         }
     }
+    aluInitResamplers();
 
     str = getenv("ALSOFT_TRAP_ERROR");
     if(str && (strcasecmp(str, "true") == 0 || strtol(str, NULL, 0) == 1))

Alc/ALu.c

@@ -82,7 +82,7 @@ extern inline ALuint64 maxu64(ALuint64 a, ALuint64 b);
 extern inline ALuint64 clampu64(ALuint64 val, ALuint64 min, ALuint64 max);
 
 extern inline ALfloat lerp(ALfloat val1, ALfloat val2, ALfloat mu);
-extern inline ALfloat cubic(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALfloat mu);
+extern inline ALfloat cubic(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALuint frac);
 
 
 static inline HrtfMixerFunc SelectHrtfMixer(void)

Alc/mixer.c

@@ -40,6 +40,23 @@
 
 extern inline void InitiatePositionArrays(ALuint frac, ALuint increment, ALuint *frac_arr, ALuint *pos_arr, ALuint size);
 
+alignas(16) ALfloat CubicLUT[FRACTIONONE][4];
+
+
+void aluInitResamplers(void)
+{
+    ALuint i;
+    for(i = 0;i < FRACTIONONE;i++)
+    {
+        ALfloat mu = (ALfloat)i / FRACTIONONE;
+        ALfloat mu2 = mu*mu, mu3 = mu*mu*mu;
+        CubicLUT[i][0] = -0.5f*mu3 +       mu2 + -0.5f*mu;
+        CubicLUT[i][1] =  1.5f*mu3 + -2.5f*mu2            + 1.0f;
+        CubicLUT[i][2] = -1.5f*mu3 +  2.0f*mu2 +  0.5f*mu;
+        CubicLUT[i][3] =  0.5f*mu3 + -0.5f*mu2;
+    }
+}
+
 
 static inline HrtfMixerFunc SelectHrtfMixer(void)
 {

Alc/mixer_c.c

@@ -13,7 +13,7 @@ static inline ALfloat point32(const ALfloat *vals, ALuint UNUSED(frac))
 static inline ALfloat lerp32(const ALfloat *vals, ALuint frac)
 { return lerp(vals[0], vals[1], frac * (1.0f/FRACTIONONE)); }
 static inline ALfloat cubic32(const ALfloat *vals, ALuint frac)
-{ return cubic(vals[-1], vals[0], vals[1], vals[2], frac * (1.0f/FRACTIONONE)); }
+{ return cubic(vals[-1], vals[0], vals[1], vals[2], frac); }
 
 const ALfloat *Resample_copy32_C(const ALfloat *src, ALuint UNUSED(frac),
   ALuint increment, ALfloat *restrict dst, ALuint numsamples)

OpenAL32/Include/alu.h

@@ -116,7 +116,7 @@ typedef void (*HrtfMixerFunc)(ALfloat (*restrict OutBuffer)[BUFFERSIZE], const A
 #define SPEEDOFSOUNDMETRESPERSEC  (343.3f)
 #define AIRABSORBGAINHF           (0.99426f) /* -0.05dB */
 
-#define FRACTIONBITS (14)
+#define FRACTIONBITS (12)
 #define FRACTIONONE  (1<<FRACTIONBITS)
 #define FRACTIONMASK (FRACTIONONE-1)
 
@@ -164,22 +164,22 @@ inline ALuint64 clampu64(ALuint64 val, ALuint64 min, ALuint64 max)
 { return minu64(max, maxu64(min, val)); }
 
 
+extern ALfloat CubicLUT[FRACTIONONE][4];
+
+
 inline ALfloat lerp(ALfloat val1, ALfloat val2, ALfloat mu)
 {
     return val1 + (val2-val1)*mu;
 }
-inline ALfloat cubic(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALfloat mu)
+inline ALfloat cubic(ALfloat val0, ALfloat val1, ALfloat val2, ALfloat val3, ALuint frac)
 {
-    ALfloat mu2 = mu*mu, mu3 = mu*mu*mu;
-    ALfloat a0 = -0.5f*mu3 +       mu2 + -0.5f*mu;
-    ALfloat a1 =  1.5f*mu3 + -2.5f*mu2            + 1.0f;
-    ALfloat a2 = -1.5f*mu3 +  2.0f*mu2 +  0.5f*mu;
-    ALfloat a3 =  0.5f*mu3 + -0.5f*mu2;
-
-    return a0*val0 + a1*val1 + a2*val2 + a3*val3;
+    const ALfloat *k = CubicLUT[frac];
+    return k[0]*val0 + k[1]*val1 + k[2]*val2 + k[3]*val3;
 }
 
 
+void aluInitResamplers(void);
+
 ALvoid aluInitPanning(ALCdevice *Device);
 
 /**