Move mixer sources into a sub-directory

8 files changed, 1100 insertions, 0 deletions

diff --git a/Alc/mixer/defs.h b/Alc/mixer/defs.h
new file mode 100644
index 00000000..fe19cef4
--- /dev/null
+++ b/Alc/mixer/defs.h
@@ -0,0 +1,119 @@
+#ifndef MIXER_DEFS_H
+#define MIXER_DEFS_H
+
+#include "AL/alc.h"
+#include "AL/al.h"
+#include "alMain.h"
+#include "alu.h"
+
+struct MixGains;
+
+struct MixHrtfParams;
+struct HrtfState;
+
+/* C resamplers */
+const ALfloat *Resample_copy_C(const InterpState *state, const ALfloat *restrict src, ALsizei frac, ALint increment, ALfloat *restrict dst, ALsizei dstlen);
+const ALfloat *Resample_point_C(const InterpState *state, const ALfloat *restrict src, ALsizei frac, ALint increment, ALfloat *restrict dst, ALsizei dstlen);
+const ALfloat *Resample_lerp_C(const InterpState *state, const ALfloat *restrict src, ALsizei frac, ALint increment, ALfloat *restrict dst, ALsizei dstlen);
+const ALfloat *Resample_cubic_C(const InterpState *state, const ALfloat *restrict src, ALsizei frac, ALint increment, ALfloat *restrict dst, ALsizei dstlen);
+const ALfloat *Resample_bsinc_C(const InterpState *state, const ALfloat *restrict src, ALsizei frac, ALint increment, ALfloat *restrict dst, ALsizei dstlen);
+
+
+/* C mixers */
+void MixHrtf_C(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+               const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+               const ALsizei IrSize, struct MixHrtfParams *hrtfparams,
+               struct HrtfState *hrtfstate, ALsizei BufferSize);
+void MixHrtfBlend_C(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                    const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+                    const ALsizei IrSize, const HrtfParams *oldparams,
+                    MixHrtfParams *newparams, HrtfState *hrtfstate,
+                    ALsizei BufferSize);
+void MixDirectHrtf_C(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                     const ALfloat *data, ALsizei Offset, const ALsizei IrSize,
+                     const ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Values)[2],
+                     ALsizei BufferSize);
+void Mix_C(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
+           ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
+           ALsizei BufferSize);
+void MixRow_C(ALfloat *OutBuffer, const ALfloat *Gains,
+              const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans,
+              ALsizei InPos, ALsizei BufferSize);
+
+/* SSE mixers */
+void MixHrtf_SSE(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                 const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+                 const ALsizei IrSize, struct MixHrtfParams *hrtfparams,
+                 struct HrtfState *hrtfstate, ALsizei BufferSize);
+void MixHrtfBlend_SSE(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                      const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+                      const ALsizei IrSize, const HrtfParams *oldparams,
+                      MixHrtfParams *newparams, HrtfState *hrtfstate,
+                      ALsizei BufferSize);
+void MixDirectHrtf_SSE(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                       const ALfloat *data, ALsizei Offset, const ALsizei IrSize,
+                       const ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Values)[2],
+                       ALsizei BufferSize);
+void Mix_SSE(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
+             ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
+             ALsizei BufferSize);
+void MixRow_SSE(ALfloat *OutBuffer, const ALfloat *Gains,
+                const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans,
+                ALsizei InPos, ALsizei BufferSize);
+
+/* SSE resamplers */
+inline void InitiatePositionArrays(ALsizei frac, ALint increment, ALsizei *restrict frac_arr, ALint *restrict pos_arr, ALsizei size)
+{
+    ALsizei i;
+
+    pos_arr[0] = 0;
+    frac_arr[0] = frac;
+    for(i = 1;i < size;i++)
+    {
+        ALint frac_tmp = frac_arr[i-1] + increment;
+        pos_arr[i] = pos_arr[i-1] + (frac_tmp>>FRACTIONBITS);
+        frac_arr[i] = frac_tmp&FRACTIONMASK;
+    }
+}
+
+const ALfloat *Resample_lerp_SSE2(const InterpState *state, const ALfloat *restrict src,
+                                  ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                  ALsizei numsamples);
+const ALfloat *Resample_lerp_SSE41(const InterpState *state, const ALfloat *restrict src,
+                                   ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                   ALsizei numsamples);
+
+const ALfloat *Resample_bsinc_SSE(const InterpState *state, const ALfloat *restrict src,
+                                  ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                  ALsizei dstlen);
+
+/* Neon mixers */
+void MixHrtf_Neon(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                  const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+                  const ALsizei IrSize, struct MixHrtfParams *hrtfparams,
+                  struct HrtfState *hrtfstate, ALsizei BufferSize);
+void MixHrtfBlend_Neon(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                       const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+                       const ALsizei IrSize, const HrtfParams *oldparams,
+                       MixHrtfParams *newparams, HrtfState *hrtfstate,
+                       ALsizei BufferSize);
+void MixDirectHrtf_Neon(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                        const ALfloat *data, ALsizei Offset, const ALsizei IrSize,
+                        const ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Values)[2],
+                        ALsizei BufferSize);
+void Mix_Neon(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
+              ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
+              ALsizei BufferSize);
+void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Gains,
+                 const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans,
+                 ALsizei InPos, ALsizei BufferSize);
+
+/* Neon resamplers */
+const ALfloat *Resample_lerp_Neon(const InterpState *state, const ALfloat *restrict src,
+                                  ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                  ALsizei numsamples);
+const ALfloat *Resample_bsinc_Neon(const InterpState *state, const ALfloat *restrict src,
+                                   ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                   ALsizei dstlen);
+
+#endif /* MIXER_DEFS_H */
diff --git a/Alc/mixer/mixer_c.c b/Alc/mixer/mixer_c.c
new file mode 100644
index 00000000..0c33e9b0
--- /dev/null
+++ b/Alc/mixer/mixer_c.c
@@ -0,0 +1,209 @@
+#include "config.h"
+
+#include <assert.h>
+
+#include "alMain.h"
+#include "alu.h"
+#include "alSource.h"
+#include "alAuxEffectSlot.h"
+#include "defs.h"
+
+
+static inline ALfloat do_point(const ALfloat *restrict vals, ALsizei UNUSED(frac))
+{ return vals[0]; }
+static inline ALfloat do_lerp(const ALfloat *restrict vals, ALsizei frac)
+{ return lerp(vals[0], vals[1], frac * (1.0f/FRACTIONONE)); }
+static inline ALfloat do_cubic(const ALfloat *restrict vals, ALsizei frac)
+{ return cubic(vals[0], vals[1], vals[2], vals[3], frac * (1.0f/FRACTIONONE)); }
+
+const ALfloat *Resample_copy_C(const InterpState* UNUSED(state),
+  const ALfloat *restrict src, ALsizei UNUSED(frac), ALint UNUSED(increment),
+  ALfloat *restrict dst, ALsizei numsamples)
+{
+#if defined(HAVE_SSE) || defined(HAVE_NEON)
+    /* Avoid copying the source data if it's aligned like the destination. */
+    if((((intptr_t)src)&15) == (((intptr_t)dst)&15))
+        return src;
+#endif
+    memcpy(dst, src, numsamples*sizeof(ALfloat));
+    return dst;
+}
+
+#define DECL_TEMPLATE(Tag, Sampler, O)                                        \
+const ALfloat *Resample_##Tag##_C(const InterpState* UNUSED(state),           \
+  const ALfloat *restrict src, ALsizei frac, ALint increment,                 \
+  ALfloat *restrict dst, ALsizei numsamples)                                  \
+{                                                                             \
+    ALsizei i;                                                                \
+                                                                              \
+    src -= O;                                                                 \
+    for(i = 0;i < numsamples;i++)                                             \
+    {                                                                         \
+        dst[i] = Sampler(src, frac);                                          \
+                                                                              \
+        frac += increment;                                                    \
+        src  += frac>>FRACTIONBITS;                                           \
+        frac &= FRACTIONMASK;                                                 \
+    }                                                                         \
+    return dst;                                                               \
+}
+
+DECL_TEMPLATE(point, do_point, 0)
+DECL_TEMPLATE(lerp, do_lerp, 0)
+DECL_TEMPLATE(cubic, do_cubic, 1)
+
+#undef DECL_TEMPLATE
+
+const ALfloat *Resample_bsinc_C(const InterpState *state, const ALfloat *restrict src,
+                                ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                ALsizei dstlen)
+{
+    const ALfloat *fil, *scd, *phd, *spd;
+    const ALfloat *const filter = state->bsinc.filter;
+    const ALfloat sf = state->bsinc.sf;
+    const ALsizei m = state->bsinc.m;
+    ALsizei j_f, pi, i;
+    ALfloat pf, r;
+
+    src += state->bsinc.l;
+    for(i = 0;i < dstlen;i++)
+    {
+        // Calculate the phase index and factor.
+#define FRAC_PHASE_BITDIFF (FRACTIONBITS-BSINC_PHASE_BITS)
+        pi = frac >> FRAC_PHASE_BITDIFF;
+        pf = (frac & ((1<<FRAC_PHASE_BITDIFF)-1)) * (1.0f/(1<<FRAC_PHASE_BITDIFF));
+#undef FRAC_PHASE_BITDIFF
+
+        fil = ASSUME_ALIGNED(filter + m*pi*4, 16);
+        scd = ASSUME_ALIGNED(fil + m, 16);
+        phd = ASSUME_ALIGNED(scd + m, 16);
+        spd = ASSUME_ALIGNED(phd + m, 16);
+
+        // Apply the scale and phase interpolated filter.
+        r = 0.0f;
+        for(j_f = 0;j_f < m;j_f++)
+            r += (fil[j_f] + sf*scd[j_f] + pf*(phd[j_f] + sf*spd[j_f])) * src[j_f];
+        dst[i] = r;
+
+        frac += increment;
+        src  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
+    }
+    return dst;
+}
+
+
+void ALfilterState_processC(ALfilterState *filter, ALfloat *restrict dst, const ALfloat *restrict src, ALsizei numsamples)
+{
+    ALsizei i;
+    if(LIKELY(numsamples > 1))
+    {
+        ALfloat x0 = filter->x[0];
+        ALfloat x1 = filter->x[1];
+        ALfloat y0 = filter->y[0];
+        ALfloat y1 = filter->y[1];
+
+        for(i = 0;i < numsamples;i++)
+        {
+            dst[i] = filter->b0* src[i] +
+                     filter->b1*x0 + filter->b2*x1 -
+                     filter->a1*y0 - filter->a2*y1;
+            y1 = y0; y0 = dst[i];
+            x1 = x0; x0 = src[i];
+        }
+
+        filter->x[0] = x0;
+        filter->x[1] = x1;
+        filter->y[0] = y0;
+        filter->y[1] = y1;
+    }
+    else if(numsamples == 1)
+    {
+        dst[0] = filter->b0 * src[0] +
+                 filter->b1 * filter->x[0] +
+                 filter->b2 * filter->x[1] -
+                 filter->a1 * filter->y[0] -
+                 filter->a2 * filter->y[1];
+        filter->x[1] = filter->x[0];
+        filter->x[0] = src[0];
+        filter->y[1] = filter->y[0];
+        filter->y[0] = dst[0];
+    }
+}
+
+
+static inline void ApplyCoeffs(ALsizei Offset, ALfloat (*restrict Values)[2],
+                               const ALsizei IrSize,
+                               const ALfloat (*restrict Coeffs)[2],
+                               ALfloat left, ALfloat right)
+{
+    ALsizei c;
+    for(c = 0;c < IrSize;c++)
+    {
+        const ALsizei off = (Offset+c)&HRIR_MASK;
+        Values[off][0] += Coeffs[c][0] * left;
+        Values[off][1] += Coeffs[c][1] * right;
+    }
+}
+
+#define MixHrtf MixHrtf_C
+#define MixHrtfBlend MixHrtfBlend_C
+#define MixDirectHrtf MixDirectHrtf_C
+#include "mixer_inc.c"
+#undef MixHrtf
+
+
+void Mix_C(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
+           ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
+           ALsizei BufferSize)
+{
+    ALfloat gain, delta, step;
+    ALsizei c;
+
+    delta = (Counter > 0) ? 1.0f/(ALfloat)Counter : 0.0f;
+
+    for(c = 0;c < OutChans;c++)
+    {
+        ALsizei pos = 0;
+        gain = CurrentGains[c];
+        step = (TargetGains[c] - gain) * delta;
+        if(fabsf(step) > FLT_EPSILON)
+        {
+            ALsizei minsize = mini(BufferSize, Counter);
+            for(;pos < minsize;pos++)
+            {
+                OutBuffer[c][OutPos+pos] += data[pos]*gain;
+                gain += step;
+            }
+            if(pos == Counter)
+                gain = TargetGains[c];
+            CurrentGains[c] = gain;
+        }
+
+        if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+            continue;
+        for(;pos < BufferSize;pos++)
+            OutBuffer[c][OutPos+pos] += data[pos]*gain;
+    }
+}
+
+/* Basically the inverse of the above. Rather than one input going to multiple
+ * outputs (each with its own gain), it's multiple inputs (each with its own
+ * gain) going to one output. This applies one row (vs one column) of a matrix
+ * transform. And as the matrices are more or less static once set up, no
+ * stepping is necessary.
+ */
+void MixRow_C(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans, ALsizei InPos, ALsizei BufferSize)
+{
+    ALsizei c, i;
+
+    for(c = 0;c < InChans;c++)
+    {
+        ALfloat gain = Gains[c];
+        if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+            continue;
+
+        for(i = 0;i < BufferSize;i++)
+            OutBuffer[i] += data[c][InPos+i] * gain;
+    }
+}
diff --git a/Alc/mixer/mixer_inc.c b/Alc/mixer/mixer_inc.c
new file mode 100644
index 00000000..ad0daa63
--- /dev/null
+++ b/Alc/mixer/mixer_inc.c
@@ -0,0 +1,114 @@
+#include "config.h"
+
+#include "alMain.h"
+#include "alSource.h"
+
+#include "hrtf.h"
+#include "align.h"
+#include "alu.h"
+#include "defs.h"
+
+
+static inline void ApplyCoeffs(ALsizei Offset, ALfloat (*restrict Values)[2],
+                               const ALsizei irSize,
+                               const ALfloat (*restrict Coeffs)[2],
+                               ALfloat left, ALfloat right);
+
+
+void MixHrtf(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+             const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+             const ALsizei IrSize, MixHrtfParams *hrtfparams, HrtfState *hrtfstate,
+             ALsizei BufferSize)
+{
+    const ALfloat (*Coeffs)[2] = ASSUME_ALIGNED(hrtfparams->Coeffs, 16);
+    const ALsizei Delay[2] = { hrtfparams->Delay[0], hrtfparams->Delay[1] };
+    ALfloat gainstep = hrtfparams->GainStep;
+    ALfloat gain = hrtfparams->Gain;
+    ALfloat left, right;
+    ALsizei i;
+
+    LeftOut  += OutPos;
+    RightOut += OutPos;
+    for(i = 0;i < BufferSize;i++)
+    {
+        hrtfstate->History[Offset&HRTF_HISTORY_MASK] = *(data++);
+        left = hrtfstate->History[(Offset-Delay[0])&HRTF_HISTORY_MASK]*gain;
+        right = hrtfstate->History[(Offset-Delay[1])&HRTF_HISTORY_MASK]*gain;
+
+        hrtfstate->Values[(Offset+IrSize-1)&HRIR_MASK][0] = 0.0f;
+        hrtfstate->Values[(Offset+IrSize-1)&HRIR_MASK][1] = 0.0f;
+
+        ApplyCoeffs(Offset, hrtfstate->Values, IrSize, Coeffs, left, right);
+        *(LeftOut++)  += hrtfstate->Values[Offset&HRIR_MASK][0];
+        *(RightOut++) += hrtfstate->Values[Offset&HRIR_MASK][1];
+
+        gain += gainstep;
+        Offset++;
+    }
+    hrtfparams->Gain = gain;
+}
+
+void MixHrtfBlend(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                  const ALfloat *data, ALsizei Offset, ALsizei OutPos,
+                  const ALsizei IrSize, const HrtfParams *oldparams,
+                  MixHrtfParams *newparams, HrtfState *hrtfstate,
+                  ALsizei BufferSize)
+{
+    const ALfloat (*OldCoeffs)[2] = ASSUME_ALIGNED(oldparams->Coeffs, 16);
+    const ALsizei OldDelay[2] = { oldparams->Delay[0], oldparams->Delay[1] };
+    ALfloat oldGain = oldparams->Gain;
+    ALfloat oldGainStep = -oldGain / (ALfloat)BufferSize;
+    const ALfloat (*NewCoeffs)[2] = ASSUME_ALIGNED(newparams->Coeffs, 16);
+    const ALsizei NewDelay[2] = { newparams->Delay[0], newparams->Delay[1] };
+    ALfloat newGain = newparams->Gain;
+    ALfloat newGainStep = newparams->GainStep;
+    ALfloat left, right;
+    ALsizei i;
+
+    LeftOut  += OutPos;
+    RightOut += OutPos;
+    for(i = 0;i < BufferSize;i++)
+    {
+        hrtfstate->Values[(Offset+IrSize-1)&HRIR_MASK][0] = 0.0f;
+        hrtfstate->Values[(Offset+IrSize-1)&HRIR_MASK][1] = 0.0f;
+
+        hrtfstate->History[Offset&HRTF_HISTORY_MASK] = *(data++);
+
+        left = hrtfstate->History[(Offset-OldDelay[0])&HRTF_HISTORY_MASK]*oldGain;
+        right = hrtfstate->History[(Offset-OldDelay[1])&HRTF_HISTORY_MASK]*oldGain;
+        ApplyCoeffs(Offset, hrtfstate->Values, IrSize, OldCoeffs, left, right);
+
+        left = hrtfstate->History[(Offset-NewDelay[0])&HRTF_HISTORY_MASK]*newGain;
+        right = hrtfstate->History[(Offset-NewDelay[1])&HRTF_HISTORY_MASK]*newGain;
+        ApplyCoeffs(Offset, hrtfstate->Values, IrSize, NewCoeffs, left, right);
+
+        *(LeftOut++)  += hrtfstate->Values[Offset&HRIR_MASK][0];
+        *(RightOut++) += hrtfstate->Values[Offset&HRIR_MASK][1];
+
+        oldGain += oldGainStep;
+        newGain += newGainStep;
+        Offset++;
+    }
+    newparams->Gain = newGain;
+}
+
+void MixDirectHrtf(ALfloat *restrict LeftOut, ALfloat *restrict RightOut,
+                   const ALfloat *data, ALsizei Offset, const ALsizei IrSize,
+                   const ALfloat (*restrict Coeffs)[2], ALfloat (*restrict Values)[2],
+                   ALsizei BufferSize)
+{
+    ALfloat insample;
+    ALsizei i;
+
+    for(i = 0;i < BufferSize;i++)
+    {
+        Values[(Offset+IrSize)&HRIR_MASK][0] = 0.0f;
+        Values[(Offset+IrSize)&HRIR_MASK][1] = 0.0f;
+        Offset++;
+
+        insample = *(data++);
+        ApplyCoeffs(Offset, Values, IrSize, Coeffs, insample, insample);
+        *(LeftOut++)  += Values[Offset&HRIR_MASK][0];
+        *(RightOut++) += Values[Offset&HRIR_MASK][1];
+    }
+}
diff --git a/Alc/mixer/mixer_neon.c b/Alc/mixer/mixer_neon.c
new file mode 100644
index 00000000..b93d11fd
--- /dev/null
+++ b/Alc/mixer/mixer_neon.c
@@ -0,0 +1,261 @@
+#include "config.h"
+
+#include <arm_neon.h>
+
+#include "AL/al.h"
+#include "AL/alc.h"
+#include "alMain.h"
+#include "alu.h"
+#include "hrtf.h"
+#include "defs.h"
+
+
+const ALfloat *Resample_lerp_Neon(const InterpState* UNUSED(state),
+  const ALfloat *restrict src, ALsizei frac, ALint increment,
+  ALfloat *restrict dst, ALsizei numsamples)
+{
+    const int32x4_t increment4 = vdupq_n_s32(increment*4);
+    const float32x4_t fracOne4 = vdupq_n_f32(1.0f/FRACTIONONE);
+    const int32x4_t fracMask4 = vdupq_n_s32(FRACTIONMASK);
+    alignas(16) ALint pos_[4];
+    alignas(16) ALsizei frac_[4];
+    int32x4_t pos4;
+    int32x4_t frac4;
+    ALsizei i;
+
+    InitiatePositionArrays(frac, increment, frac_, pos_, 4);
+
+    frac4 = vld1q_s32(frac_);
+    pos4 = vld1q_s32(pos_);
+
+    for(i = 0;numsamples-i > 3;i += 4)
+    {
+        const float32x4_t val1 = (float32x4_t){src[pos_[0]], src[pos_[1]], src[pos_[2]], src[pos_[3]]};
+        const float32x4_t val2 = (float32x4_t){src[pos_[0]+1], src[pos_[1]+1], src[pos_[2]+1], src[pos_[3]+1]};
+
+        /* val1 + (val2-val1)*mu */
+        const float32x4_t r0 = vsubq_f32(val2, val1);
+        const float32x4_t mu = vmulq_f32(vcvtq_f32_s32(frac4), fracOne4);
+        const float32x4_t out = vmlaq_f32(val1, mu, r0);
+
+        vst1q_f32(&dst[i], out);
+
+        frac4 = vaddq_s32(frac4, increment4);
+        pos4 = vaddq_s32(pos4, vshrq_n_s32(frac4, FRACTIONBITS));
+        frac4 = vandq_s32(frac4, fracMask4);
+
+        vst1q_s32(pos_, pos4);
+    }
+
+    if(i < numsamples)
+    {
+        /* NOTE: These four elements represent the position *after* the last
+         * four samples, so the lowest element is the next position to
+         * resample.
+         */
+        ALint pos = pos_[0];
+        frac = vgetq_lane_s32(frac4, 0);
+        do {
+            dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
+
+            frac += increment;
+            pos  += frac>>FRACTIONBITS;
+            frac &= FRACTIONMASK;
+        } while(++i < numsamples);
+    }
+    return dst;
+}
+
+const ALfloat *Resample_bsinc_Neon(const InterpState *state,
+  const ALfloat *restrict src, ALsizei frac, ALint increment,
+  ALfloat *restrict dst, ALsizei dstlen)
+{
+    const ALfloat *const filter = state->bsinc.filter;
+    const float32x4_t sf4 = vdupq_n_f32(state->bsinc.sf);
+    const ALsizei m = state->bsinc.m;
+    const float32x4_t *fil, *scd, *phd, *spd;
+    ALsizei pi, i, j, offset;
+    float32x4_t r4;
+    ALfloat pf;
+
+    src += state->bsinc.l;
+    for(i = 0;i < dstlen;i++)
+    {
+        // Calculate the phase index and factor.
+#define FRAC_PHASE_BITDIFF (FRACTIONBITS-BSINC_PHASE_BITS)
+        pi = frac >> FRAC_PHASE_BITDIFF;
+        pf = (frac & ((1<<FRAC_PHASE_BITDIFF)-1)) * (1.0f/(1<<FRAC_PHASE_BITDIFF));
+#undef FRAC_PHASE_BITDIFF
+
+        offset = m*pi*4;
+        fil = ASSUME_ALIGNED(filter + offset, 16); offset += m;
+        scd = ASSUME_ALIGNED(filter + offset, 16); offset += m;
+        phd = ASSUME_ALIGNED(filter + offset, 16); offset += m;
+        spd = ASSUME_ALIGNED(filter + offset, 16);
+
+        // Apply the scale and phase interpolated filter.
+        r4 = vdupq_n_f32(0.0f);
+        {
+            const float32x4_t pf4 = vdupq_n_f32(pf);
+            for(j = 0;j < m;j+=4,fil++,scd++,phd++,spd++)
+            {
+                /* f = ((fil + sf*scd) + pf*(phd + sf*spd)) */
+                const float32x4_t f4 = vmlaq_f32(
+                    vmlaq_f32(*fil, sf4, *scd),
+                    pf4, vmlaq_f32(*phd, sf4, *spd)
+                );
+                /* r += f*src */
+                r4 = vmlaq_f32(r4, f4, vld1q_f32(&src[j]));
+            }
+        }
+        r4 = vaddq_f32(r4, vcombine_f32(vrev64_f32(vget_high_f32(r4)),
+                                        vrev64_f32(vget_low_f32(r4))));
+        dst[i] = vget_lane_f32(vadd_f32(vget_low_f32(r4), vget_high_f32(r4)), 0);
+
+        frac += increment;
+        src  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
+    }
+    return dst;
+}
+
+
+static inline void ApplyCoeffs(ALsizei Offset, ALfloat (*restrict Values)[2],
+                               const ALsizei IrSize,
+                               const ALfloat (*restrict Coeffs)[2],
+                               ALfloat left, ALfloat right)
+{
+    ALsizei c;
+    float32x4_t leftright4;
+    {
+        float32x2_t leftright2 = vdup_n_f32(0.0);
+        leftright2 = vset_lane_f32(left, leftright2, 0);
+        leftright2 = vset_lane_f32(right, leftright2, 1);
+        leftright4 = vcombine_f32(leftright2, leftright2);
+    }
+    Values = ASSUME_ALIGNED(Values, 16);
+    Coeffs = ASSUME_ALIGNED(Coeffs, 16);
+    for(c = 0;c < IrSize;c += 2)
+    {
+        const ALsizei o0 = (Offset+c)&HRIR_MASK;
+        const ALsizei o1 = (o0+1)&HRIR_MASK;
+        float32x4_t vals = vcombine_f32(vld1_f32((float32_t*)&Values[o0][0]),
+                                        vld1_f32((float32_t*)&Values[o1][0]));
+        float32x4_t coefs = vld1q_f32((float32_t*)&Coeffs[c][0]);
+
+        vals = vmlaq_f32(vals, coefs, leftright4);
+
+        vst1_f32((float32_t*)&Values[o0][0], vget_low_f32(vals));
+        vst1_f32((float32_t*)&Values[o1][0], vget_high_f32(vals));
+    }
+}
+
+#define MixHrtf MixHrtf_Neon
+#define MixHrtfBlend MixHrtfBlend_Neon
+#define MixDirectHrtf MixDirectHrtf_Neon
+#include "mixer_inc.c"
+#undef MixHrtf
+
+
+void Mix_Neon(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
+              ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
+              ALsizei BufferSize)
+{
+    ALfloat gain, delta, step;
+    float32x4_t gain4;
+    ALsizei c;
+
+    data = ASSUME_ALIGNED(data, 16);
+    OutBuffer = ASSUME_ALIGNED(OutBuffer, 16);
+
+    delta = (Counter > 0) ? 1.0f/(ALfloat)Counter : 0.0f;
+
+    for(c = 0;c < OutChans;c++)
+    {
+        ALsizei pos = 0;
+        gain = CurrentGains[c];
+        step = (TargetGains[c] - gain) * delta;
+        if(fabsf(step) > FLT_EPSILON)
+        {
+            ALsizei minsize = mini(BufferSize, Counter);
+            /* Mix with applying gain steps in aligned multiples of 4. */
+            if(minsize-pos > 3)
+            {
+                float32x4_t step4;
+                gain4 = vsetq_lane_f32(gain, gain4, 0);
+                gain4 = vsetq_lane_f32(gain + step, gain4, 1);
+                gain4 = vsetq_lane_f32(gain + step + step, gain4, 2);
+                gain4 = vsetq_lane_f32(gain + step + step + step, gain4, 3);
+                step4 = vdupq_n_f32(step + step + step + step);
+                do {
+                    const float32x4_t val4 = vld1q_f32(&data[pos]);
+                    float32x4_t dry4 = vld1q_f32(&OutBuffer[c][OutPos+pos]);
+                    dry4 = vmlaq_f32(dry4, val4, gain4);
+                    gain4 = vaddq_f32(gain4, step4);
+                    vst1q_f32(&OutBuffer[c][OutPos+pos], dry4);
+                    pos += 4;
+                } while(minsize-pos > 3);
+                /* NOTE: gain4 now represents the next four gains after the
+                 * last four mixed samples, so the lowest element represents
+                 * the next gain to apply.
+                 */
+                gain = vgetq_lane_f32(gain4, 0);
+            }
+            /* Mix with applying left over gain steps that aren't aligned multiples of 4. */
+            for(;pos < minsize;pos++)
+            {
+                OutBuffer[c][OutPos+pos] += data[pos]*gain;
+                gain += step;
+            }
+            if(pos == Counter)
+                gain = TargetGains[c];
+            CurrentGains[c] = gain;
+
+            /* Mix until pos is aligned with 4 or the mix is done. */
+            minsize = mini(BufferSize, (pos+3)&~3);
+            for(;pos < minsize;pos++)
+                OutBuffer[c][OutPos+pos] += data[pos]*gain;
+        }
+
+        if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+            continue;
+        gain4 = vdupq_n_f32(gain);
+        for(;BufferSize-pos > 3;pos += 4)
+        {
+            const float32x4_t val4 = vld1q_f32(&data[pos]);
+            float32x4_t dry4 = vld1q_f32(&OutBuffer[c][OutPos+pos]);
+            dry4 = vmlaq_f32(dry4, val4, gain4);
+            vst1q_f32(&OutBuffer[c][OutPos+pos], dry4);
+        }
+        for(;pos < BufferSize;pos++)
+            OutBuffer[c][OutPos+pos] += data[pos]*gain;
+    }
+}
+
+void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans, ALsizei InPos, ALsizei BufferSize)
+{
+    float32x4_t gain4;
+    ALsizei c;
+
+    data = ASSUME_ALIGNED(data, 16);
+    OutBuffer = ASSUME_ALIGNED(OutBuffer, 16);
+
+    for(c = 0;c < InChans;c++)
+    {
+        ALsizei pos = 0;
+        ALfloat gain = Gains[c];
+        if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+            continue;
+
+        gain4 = vdupq_n_f32(gain);
+        for(;BufferSize-pos > 3;pos += 4)
+        {
+            const float32x4_t val4 = vld1q_f32(&data[c][InPos+pos]);
+            float32x4_t dry4 = vld1q_f32(&OutBuffer[pos]);
+            dry4 = vmlaq_f32(dry4, val4, gain4);
+            vst1q_f32(&OutBuffer[pos], dry4);
+        }
+        for(;pos < BufferSize;pos++)
+            OutBuffer[pos] += data[c][InPos+pos]*gain;
+    }
+}
diff --git a/Alc/mixer/mixer_sse.c b/Alc/mixer/mixer_sse.c
new file mode 100644
index 00000000..288661b2
--- /dev/null
+++ b/Alc/mixer/mixer_sse.c
@@ -0,0 +1,229 @@
+#include "config.h"
+
+#include <xmmintrin.h>
+
+#include "AL/al.h"
+#include "AL/alc.h"
+#include "alMain.h"
+#include "alu.h"
+
+#include "alSource.h"
+#include "alAuxEffectSlot.h"
+#include "defs.h"
+
+
+const ALfloat *Resample_bsinc_SSE(const InterpState *state, const ALfloat *restrict src,
+                                  ALsizei frac, ALint increment, ALfloat *restrict dst,
+                                  ALsizei dstlen)
+{
+    const ALfloat *const filter = state->bsinc.filter;
+    const __m128 sf4 = _mm_set1_ps(state->bsinc.sf);
+    const ALsizei m = state->bsinc.m;
+    const __m128 *fil, *scd, *phd, *spd;
+    ALsizei pi, i, j, offset;
+    ALfloat pf;
+    __m128 r4;
+
+    src += state->bsinc.l;
+    for(i = 0;i < dstlen;i++)
+    {
+        // Calculate the phase index and factor.
+#define FRAC_PHASE_BITDIFF (FRACTIONBITS-BSINC_PHASE_BITS)
+        pi = frac >> FRAC_PHASE_BITDIFF;
+        pf = (frac & ((1<<FRAC_PHASE_BITDIFF)-1)) * (1.0f/(1<<FRAC_PHASE_BITDIFF));
+#undef FRAC_PHASE_BITDIFF
+
+        offset = m*pi*4;
+        fil = (const __m128*)ASSUME_ALIGNED(filter + offset, 16); offset += m;
+        scd = (const __m128*)ASSUME_ALIGNED(filter + offset, 16); offset += m;
+        phd = (const __m128*)ASSUME_ALIGNED(filter + offset, 16); offset += m;
+        spd = (const __m128*)ASSUME_ALIGNED(filter + offset, 16);
+
+        // Apply the scale and phase interpolated filter.
+        r4 = _mm_setzero_ps();
+        {
+            const __m128 pf4 = _mm_set1_ps(pf);
+#define MLA4(x, y, z) _mm_add_ps(x, _mm_mul_ps(y, z))
+            for(j = 0;j < m;j+=4,fil++,scd++,phd++,spd++)
+            {
+                /* f = ((fil + sf*scd) + pf*(phd + sf*spd)) */
+                const __m128 f4 = MLA4(
+                    MLA4(*fil, sf4, *scd),
+                    pf4, MLA4(*phd, sf4, *spd)
+                );
+                /* r += f*src */
+                r4 = MLA4(r4, f4, _mm_loadu_ps(&src[j]));
+            }
+#undef MLA4
+        }
+        r4 = _mm_add_ps(r4, _mm_shuffle_ps(r4, r4, _MM_SHUFFLE(0, 1, 2, 3)));
+        r4 = _mm_add_ps(r4, _mm_movehl_ps(r4, r4));
+        dst[i] = _mm_cvtss_f32(r4);
+
+        frac += increment;
+        src  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
+    }
+    return dst;
+}
+
+
+static inline void ApplyCoeffs(ALsizei Offset, ALfloat (*restrict Values)[2],
+                               const ALsizei IrSize,
+                               const ALfloat (*restrict Coeffs)[2],
+                               ALfloat left, ALfloat right)
+{
+    const __m128 lrlr = _mm_setr_ps(left, right, left, right);
+    __m128 vals = _mm_setzero_ps();
+    __m128 coeffs;
+    ALsizei i;
+
+    Values = ASSUME_ALIGNED(Values, 16);
+    Coeffs = ASSUME_ALIGNED(Coeffs, 16);
+    if((Offset&1))
+    {
+        const ALsizei o0 = Offset&HRIR_MASK;
+        const ALsizei o1 = (Offset+IrSize-1)&HRIR_MASK;
+        __m128 imp0, imp1;
+
+        coeffs = _mm_load_ps(&Coeffs[0][0]);
+        vals = _mm_loadl_pi(vals, (__m64*)&Values[o0][0]);
+        imp0 = _mm_mul_ps(lrlr, coeffs);
+        vals = _mm_add_ps(imp0, vals);
+        _mm_storel_pi((__m64*)&Values[o0][0], vals);
+        for(i = 1;i < IrSize-1;i += 2)
+        {
+            const ALsizei o2 = (Offset+i)&HRIR_MASK;
+
+            coeffs = _mm_load_ps(&Coeffs[i+1][0]);
+            vals = _mm_load_ps(&Values[o2][0]);
+            imp1 = _mm_mul_ps(lrlr, coeffs);
+            imp0 = _mm_shuffle_ps(imp0, imp1, _MM_SHUFFLE(1, 0, 3, 2));
+            vals = _mm_add_ps(imp0, vals);
+            _mm_store_ps(&Values[o2][0], vals);
+            imp0 = imp1;
+        }
+        vals = _mm_loadl_pi(vals, (__m64*)&Values[o1][0]);
+        imp0 = _mm_movehl_ps(imp0, imp0);
+        vals = _mm_add_ps(imp0, vals);
+        _mm_storel_pi((__m64*)&Values[o1][0], vals);
+    }
+    else
+    {
+        for(i = 0;i < IrSize;i += 2)
+        {
+            const ALsizei o = (Offset + i)&HRIR_MASK;
+
+            coeffs = _mm_load_ps(&Coeffs[i][0]);
+            vals = _mm_load_ps(&Values[o][0]);
+            vals = _mm_add_ps(vals, _mm_mul_ps(lrlr, coeffs));
+            _mm_store_ps(&Values[o][0], vals);
+        }
+    }
+}
+
+#define MixHrtf MixHrtf_SSE
+#define MixHrtfBlend MixHrtfBlend_SSE
+#define MixDirectHrtf MixDirectHrtf_SSE
+#include "mixer_inc.c"
+#undef MixHrtf
+
+
+void Mix_SSE(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
+             ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
+             ALsizei BufferSize)
+{
+    ALfloat gain, delta, step;
+    __m128 gain4;
+    ALsizei c;
+
+    delta = (Counter > 0) ? 1.0f/(ALfloat)Counter : 0.0f;
+
+    for(c = 0;c < OutChans;c++)
+    {
+        ALsizei pos = 0;
+        gain = CurrentGains[c];
+        step = (TargetGains[c] - gain) * delta;
+        if(fabsf(step) > FLT_EPSILON)
+        {
+            ALsizei minsize = mini(BufferSize, Counter);
+            /* Mix with applying gain steps in aligned multiples of 4. */
+            if(minsize-pos > 3)
+            {
+                __m128 step4;
+                gain4 = _mm_setr_ps(
+                    gain,
+                    gain + step,
+                    gain + step + step,
+                    gain + step + step + step
+                );
+                step4 = _mm_set1_ps(step + step + step + step);
+                do {
+                    const __m128 val4 = _mm_load_ps(&data[pos]);
+                    __m128 dry4 = _mm_load_ps(&OutBuffer[c][OutPos+pos]);
+                    dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
+                    gain4 = _mm_add_ps(gain4, step4);
+                    _mm_store_ps(&OutBuffer[c][OutPos+pos], dry4);
+                    pos += 4;
+                } while(minsize-pos > 3);
+                /* NOTE: gain4 now represents the next four gains after the
+                 * last four mixed samples, so the lowest element represents
+                 * the next gain to apply.
+                 */
+                gain = _mm_cvtss_f32(gain4);
+            }
+            /* Mix with applying left over gain steps that aren't aligned multiples of 4. */
+            for(;pos < minsize;pos++)
+            {
+                OutBuffer[c][OutPos+pos] += data[pos]*gain;
+                gain += step;
+            }
+            if(pos == Counter)
+                gain = TargetGains[c];
+            CurrentGains[c] = gain;
+
+            /* Mix until pos is aligned with 4 or the mix is done. */
+            minsize = mini(BufferSize, (pos+3)&~3);
+            for(;pos < minsize;pos++)
+                OutBuffer[c][OutPos+pos] += data[pos]*gain;
+        }
+
+        if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+            continue;
+        gain4 = _mm_set1_ps(gain);
+        for(;BufferSize-pos > 3;pos += 4)
+        {
+            const __m128 val4 = _mm_load_ps(&data[pos]);
+            __m128 dry4 = _mm_load_ps(&OutBuffer[c][OutPos+pos]);
+            dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
+            _mm_store_ps(&OutBuffer[c][OutPos+pos], dry4);
+        }
+        for(;pos < BufferSize;pos++)
+            OutBuffer[c][OutPos+pos] += data[pos]*gain;
+    }
+}
+
+void MixRow_SSE(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans, ALsizei InPos, ALsizei BufferSize)
+{
+    __m128 gain4;
+    ALsizei c;
+
+    for(c = 0;c < InChans;c++)
+    {
+        ALsizei pos = 0;
+        ALfloat gain = Gains[c];
+        if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+            continue;
+
+        gain4 = _mm_set1_ps(gain);
+        for(;BufferSize-pos > 3;pos += 4)
+        {
+            const __m128 val4 = _mm_load_ps(&data[c][InPos+pos]);
+            __m128 dry4 = _mm_load_ps(&OutBuffer[pos]);
+            dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
+            _mm_store_ps(&OutBuffer[pos], dry4);
+        }
+        for(;pos < BufferSize;pos++)
+            OutBuffer[pos] += data[c][InPos+pos]*gain;
+    }
+}
diff --git a/Alc/mixer/mixer_sse2.c b/Alc/mixer/mixer_sse2.c
new file mode 100644
index 00000000..19d07719
--- /dev/null
+++ b/Alc/mixer/mixer_sse2.c
@@ -0,0 +1,82 @@
+/**
+ * OpenAL cross platform audio library
+ * Copyright (C) 2014 by Timothy Arceri <[email protected]>.
+ * This library is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU Library General Public
+ *  License as published by the Free Software Foundation; either
+ *  version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ *  Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU Library General Public
+ *  License along with this library; if not, write to the
+ *  Free Software Foundation, Inc.,
+ *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ * Or go to http://www.gnu.org/copyleft/lgpl.html
+ */
+
+#include "config.h"
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+
+#include "alu.h"
+#include "defs.h"
+
+
+const ALfloat *Resample_lerp_SSE2(const InterpState* UNUSED(state),
+  const ALfloat *restrict src, ALsizei frac, ALint increment,
+  ALfloat *restrict dst, ALsizei numsamples)
+{
+    const __m128i increment4 = _mm_set1_epi32(increment*4);
+    const __m128 fracOne4 = _mm_set1_ps(1.0f/FRACTIONONE);
+    const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
+    union { alignas(16) ALint i[4]; float f[4]; } pos_;
+    union { alignas(16) ALsizei i[4]; float f[4]; } frac_;
+    __m128i frac4, pos4;
+    ALint pos;
+    ALsizei i;
+
+    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));
+
+    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]);
+
+        /* val1 + (val2-val1)*mu */
+        const __m128 r0 = _mm_sub_ps(val2, val1);
+        const __m128 mu = _mm_mul_ps(_mm_cvtepi32_ps(frac4), fracOne4);
+        const __m128 out = _mm_add_ps(val1, _mm_mul_ps(mu, r0));
+
+        _mm_store_ps(&dst[i], out);
+
+        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));
+    }
+
+    /* 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];
+    frac = _mm_cvtsi128_si32(frac4);
+
+    for(;i < numsamples;i++)
+    {
+        dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
+
+        frac += increment;
+        pos  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
+    }
+    return dst;
+}
diff --git a/Alc/mixer/mixer_sse3.c b/Alc/mixer/mixer_sse3.c
new file mode 100644
index 00000000..e69de29b
--- /dev/null
+++ b/Alc/mixer/mixer_sse3.c
diff --git a/Alc/mixer/mixer_sse41.c b/Alc/mixer/mixer_sse41.c
new file mode 100644
index 00000000..85fd0f5e
--- /dev/null
+++ b/Alc/mixer/mixer_sse41.c
@@ -0,0 +1,86 @@
+/**
+ * OpenAL cross platform audio library
+ * Copyright (C) 2014 by Timothy Arceri <[email protected]>.
+ * This library is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU Library General Public
+ *  License as published by the Free Software Foundation; either
+ *  version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ *  Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU Library General Public
+ *  License along with this library; if not, write to the
+ *  Free Software Foundation, Inc.,
+ *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ * Or go to http://www.gnu.org/copyleft/lgpl.html
+ */
+
+#include "config.h"
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+#include <smmintrin.h>
+
+#include "alu.h"
+#include "defs.h"
+
+
+const ALfloat *Resample_lerp_SSE41(const InterpState* UNUSED(state),
+  const ALfloat *restrict src, ALsizei frac, ALint increment,
+  ALfloat *restrict dst, ALsizei numsamples)
+{
+    const __m128i increment4 = _mm_set1_epi32(increment*4);
+    const __m128 fracOne4 = _mm_set1_ps(1.0f/FRACTIONONE);
+    const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
+    union { alignas(16) ALint i[4]; float f[4]; } pos_;
+    union { alignas(16) ALsizei i[4]; float f[4]; } frac_;
+    __m128i frac4, pos4;
+    ALint pos;
+    ALsizei i;
+
+    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));
+
+    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]);
+
+        /* val1 + (val2-val1)*mu */
+        const __m128 r0 = _mm_sub_ps(val2, val1);
+        const __m128 mu = _mm_mul_ps(_mm_cvtepi32_ps(frac4), fracOne4);
+        const __m128 out = _mm_add_ps(val1, _mm_mul_ps(mu, r0));
+
+        _mm_store_ps(&dst[i], out);
+
+        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];
+    frac = _mm_cvtsi128_si32(frac4);
+
+    for(;i < numsamples;i++)
+    {
+        dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
+
+        frac += increment;
+        pos  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
+    }
+    return dst;
+}