Remove the Offset parameter from ApplyCoeffs

4 files changed, 97 insertions, 88 deletions

diff --git a/alc/mixer/hrtfbase.h b/alc/mixer/hrtfbase.h
index 58d168ed..cbc885c5 100644
--- a/alc/mixer/hrtfbase.h
+++ b/alc/mixer/hrtfbase.h
@@ -9,10 +9,10 @@
 #include "voice.h"
 
 
-using ApplyCoeffsT = void(size_t Offset, float2 *RESTRICT Values, const ALuint irSize,
-    const HrirArray &Coeffs, const float left, const float right);
+using ApplyCoeffsT = void(&)(float2 *RESTRICT Values, const ALuint irSize, const HrirArray &Coeffs,
+    const float left, const float right);
 
-template<ApplyCoeffsT &ApplyCoeffs>
+template<ApplyCoeffsT ApplyCoeffs>
 inline void MixHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     const float *InSamples, float2 *RESTRICT AccumSamples, const size_t OutPos,
     const ALuint IrSize, MixHrtfFilter *hrtfparams, const size_t BufferSize)
@@ -33,7 +33,7 @@ inline void MixHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
         const float g{gain + gainstep*stepcount};
         const float left{InSamples[Delay[0]++] * g};
         const float right{InSamples[Delay[1]++] * g};
-        ApplyCoeffs(i, AccumSamples+i, IrSize, Coeffs, left, right);
+        ApplyCoeffs(AccumSamples+i, IrSize, Coeffs, left, right);
 
         stepcount += 1.0f;
     }
@@ -46,7 +46,7 @@ inline void MixHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     hrtfparams->Gain = gain + gainstep*stepcount;
 }
 
-template<ApplyCoeffsT &ApplyCoeffs>
+template<ApplyCoeffsT ApplyCoeffs>
 inline void MixHrtfBlendBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     const float *InSamples, float2 *RESTRICT AccumSamples, const size_t OutPos,
     const ALuint IrSize, const HrtfFilter *oldparams, MixHrtfFilter *newparams,
@@ -70,7 +70,7 @@ inline void MixHrtfBlendBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut
         const float g{oldGain + oldGainStep*stepcount};
         const float left{InSamples[Delay[0]++] * g};
         const float right{InSamples[Delay[1]++] * g};
-        ApplyCoeffs(i, AccumSamples+i, IrSize, OldCoeffs, left, right);
+        ApplyCoeffs(AccumSamples+i, IrSize, OldCoeffs, left, right);
 
         stepcount += 1.0f;
     }
@@ -84,7 +84,7 @@ inline void MixHrtfBlendBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut
         const float g{newGainStep*stepcount};
         const float left{InSamples[Delay[0]++] * g};
         const float right{InSamples[Delay[1]++] * g};
-        ApplyCoeffs(i, AccumSamples+i, IrSize, NewCoeffs, left, right);
+        ApplyCoeffs(AccumSamples+i, IrSize, NewCoeffs, left, right);
 
         stepcount += 1.0f;
     }
@@ -97,7 +97,7 @@ inline void MixHrtfBlendBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut
     newparams->Gain = newGainStep*stepcount;
 }
 
-template<ApplyCoeffsT &ApplyCoeffs>
+template<ApplyCoeffsT ApplyCoeffs>
 inline void MixDirectHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     const al::span<const FloatBufferLine> InSamples, float2 *RESTRICT AccumSamples,
     DirectHrtfState *State, const size_t BufferSize)
@@ -116,7 +116,7 @@ inline void MixDirectHrtfBase(FloatBufferLine &LeftOut, FloatBufferLine &RightOu
         for(size_t i{0u};i < BufferSize;++i)
         {
             const float insample{input[i]};
-            ApplyCoeffs(i, AccumSamples+i, IrSize, Coeffs, insample, insample);
+            ApplyCoeffs(AccumSamples+i, IrSize, Coeffs, insample, insample);
         }
     }
     for(size_t i{0u};i < BufferSize;++i)
diff --git a/alc/mixer/mixer_c.cpp b/alc/mixer/mixer_c.cpp
index 8aa32e6b..ffc07acd 100644
--- a/alc/mixer/mixer_c.cpp
+++ b/alc/mixer/mixer_c.cpp
@@ -68,7 +68,7 @@ const float *DoResample(const InterpState *state, const float *RESTRICT src, ALu
     ALuint increment, const al::span<float> dst)
 {
     const InterpState istate{*state};
-    auto proc_sample = [&src,&frac,istate,increment]() -> ALfloat
+    auto proc_sample = [&src,&frac,istate,increment]() -> float
     {
         const float ret{Sampler(istate, src, frac)};
 
@@ -83,6 +83,17 @@ const float *DoResample(const InterpState *state, const float *RESTRICT src, ALu
     return dst.begin();
 }
 
+inline void ApplyCoeffs(float2 *RESTRICT Values, const ALuint IrSize, const HrirArray &Coeffs,
+    const float left, const float right)
+{
+    ASSUME(IrSize >= 4);
+    for(ALuint c{0};c < IrSize;++c)
+    {
+        Values[c][0] += Coeffs[c][0] * left;
+        Values[c][1] += Coeffs[c][1] * right;
+    }
+}
+
 } // namespace
 
 template<>
@@ -124,17 +135,6 @@ const ALfloat *Resample_<FastBSincTag,CTag>(const InterpState *state, const ALfl
 { return DoResample<do_fastbsinc>(state, src-state->bsinc.l, frac, increment, dst); }
 
 
-static inline void ApplyCoeffs(size_t /*Offset*/, float2 *RESTRICT Values, const ALuint IrSize,
-    const HrirArray &Coeffs, const float left, const float right)
-{
-    ASSUME(IrSize >= 4);
-    for(ALuint c{0};c < IrSize;++c)
-    {
-        Values[c][0] += Coeffs[c][0] * left;
-        Values[c][1] += Coeffs[c][1] * right;
-    }
-}
-
 template<>
 void MixHrtf_<CTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     const float *InSamples, float2 *AccumSamples, const size_t OutPos, const ALuint IrSize,
diff --git a/alc/mixer/mixer_neon.cpp b/alc/mixer/mixer_neon.cpp
index c0fc1651..ae782897 100644
--- a/alc/mixer/mixer_neon.cpp
+++ b/alc/mixer/mixer_neon.cpp
@@ -13,6 +13,32 @@
 #include "hrtfbase.h"
 
 
+namespace {
+
+inline void ApplyCoeffs(float2 *RESTRICT Values, const ALuint IrSize, const HrirArray &Coeffs,
+    const float left, const float right)
+{
+    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);
+    }
+
+    ASSUME(IrSize >= 4);
+    for(ALuint c{0};c < IrSize;c += 2)
+    {
+        float32x4_t vals = vld1q_f32(&Values[c][0]);
+        float32x4_t coefs = vld1q_f32(&Coeffs[c][0]);
+
+        vals = vmlaq_f32(vals, coefs, leftright4);
+
+        vst1q_f32(&Values[c][0], vals);
+    }
+}
+
+} // namespace
 
 template<>
 const ALfloat *Resample_<LerpTag,NEONTag>(const InterpState*, const ALfloat *RESTRICT src,
@@ -163,30 +189,6 @@ const ALfloat *Resample_<FastBSincTag,NEONTag>(const InterpState *state,
 }
 
 
-static inline void ApplyCoeffs(size_t /*Offset*/, float2 *RESTRICT Values, const ALuint IrSize,
-    const HrirArray &Coeffs, const float left, const float right)
-{
-    ASSUME(IrSize >= 4);
-
-    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);
-    }
-
-    for(ALuint c{0};c < IrSize;c += 2)
-    {
-        float32x4_t vals = vld1q_f32(&Values[c][0]);
-        float32x4_t coefs = vld1q_f32(&Coeffs[c][0]);
-
-        vals = vmlaq_f32(vals, coefs, leftright4);
-
-        vst1q_f32(&Values[c][0], vals);
-    }
-}
-
 template<>
 void MixHrtf_<NEONTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     const float *InSamples, float2 *AccumSamples, const size_t OutPos, const ALuint IrSize,
diff --git a/alc/mixer/mixer_sse.cpp b/alc/mixer/mixer_sse.cpp
index 0fb954aa..62ce5eab 100644
--- a/alc/mixer/mixer_sse.cpp
+++ b/alc/mixer/mixer_sse.cpp
@@ -13,6 +13,56 @@
 #include "hrtfbase.h"
 
 
+namespace {
+
+inline void ApplyCoeffs(float2 *RESTRICT Values, const ALuint IrSize, const HrirArray &Coeffs,
+    const float left, const float right)
+{
+    const __m128 lrlr{_mm_setr_ps(left, right, left, right)};
+
+    ASSUME(IrSize >= 4);
+    /* This isn't technically correct to test alignment, but it's true for
+     * systems that support SSE, which is the only one that needs to know the
+     * alignment of Values (which alternates between 8- and 16-byte aligned).
+     */
+    if(reinterpret_cast<intptr_t>(Values)&0x8)
+    {
+        __m128 imp0, imp1;
+        __m128 coeffs{_mm_load_ps(&Coeffs[0][0])};
+        __m128 vals{_mm_loadl_pi(_mm_setzero_ps(), reinterpret_cast<__m64*>(&Values[0][0]))};
+        imp0 = _mm_mul_ps(lrlr, coeffs);
+        vals = _mm_add_ps(imp0, vals);
+        _mm_storel_pi(reinterpret_cast<__m64*>(&Values[0][0]), vals);
+        ALuint i{1};
+        for(;i < IrSize-1;i += 2)
+        {
+            coeffs = _mm_load_ps(&Coeffs[i+1][0]);
+            vals = _mm_load_ps(&Values[i][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[i][0], vals);
+            imp0 = imp1;
+        }
+        vals = _mm_loadl_pi(vals, reinterpret_cast<__m64*>(&Values[i][0]));
+        imp0 = _mm_movehl_ps(imp0, imp0);
+        vals = _mm_add_ps(imp0, vals);
+        _mm_storel_pi(reinterpret_cast<__m64*>(&Values[i][0]), vals);
+    }
+    else
+    {
+        for(ALuint i{0};i < IrSize;i += 2)
+        {
+            __m128 coeffs{_mm_load_ps(&Coeffs[i][0])};
+            __m128 vals{_mm_load_ps(&Values[i][0])};
+            vals = _mm_add_ps(vals, _mm_mul_ps(lrlr, coeffs));
+            _mm_store_ps(&Values[i][0], vals);
+        }
+    }
+}
+
+} // namespace
+
 template<>
 const ALfloat *Resample_<BSincTag,SSETag>(const InterpState *state, const ALfloat *RESTRICT src,
     ALuint frac, ALuint increment, const al::span<float> dst)
@@ -114,49 +164,6 @@ const ALfloat *Resample_<FastBSincTag,SSETag>(const InterpState *state,
 }
 
 
-static inline void ApplyCoeffs(size_t Offset, float2 *RESTRICT Values, const ALuint IrSize,
-    const HrirArray &Coeffs, const float left, const float right)
-{
-    const __m128 lrlr{_mm_setr_ps(left, right, left, right)};
-
-    ASSUME(IrSize >= 4);
-
-    if((Offset&1))
-    {
-        __m128 imp0, imp1;
-        __m128 coeffs{_mm_load_ps(&Coeffs[0][0])};
-        __m128 vals{_mm_loadl_pi(_mm_setzero_ps(), reinterpret_cast<__m64*>(&Values[0][0]))};
-        imp0 = _mm_mul_ps(lrlr, coeffs);
-        vals = _mm_add_ps(imp0, vals);
-        _mm_storel_pi(reinterpret_cast<__m64*>(&Values[0][0]), vals);
-        ALuint i{1};
-        for(;i < IrSize-1;i += 2)
-        {
-            coeffs = _mm_load_ps(&Coeffs[i+1][0]);
-            vals = _mm_load_ps(&Values[i][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[i][0], vals);
-            imp0 = imp1;
-        }
-        vals = _mm_loadl_pi(vals, reinterpret_cast<__m64*>(&Values[i][0]));
-        imp0 = _mm_movehl_ps(imp0, imp0);
-        vals = _mm_add_ps(imp0, vals);
-        _mm_storel_pi(reinterpret_cast<__m64*>(&Values[i][0]), vals);
-    }
-    else
-    {
-        for(ALuint i{0};i < IrSize;i += 2)
-        {
-            __m128 coeffs{_mm_load_ps(&Coeffs[i][0])};
-            __m128 vals{_mm_load_ps(&Values[i][0])};
-            vals = _mm_add_ps(vals, _mm_mul_ps(lrlr, coeffs));
-            _mm_store_ps(&Values[i][0], vals);
-        }
-    }
-}
-
 template<>
 void MixHrtf_<SSETag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
     const float *InSamples, float2 *AccumSamples, const size_t OutPos, const ALuint IrSize,