11 files changed, 222 insertions, 8 deletions
diff --git a/alc/effects/autowah.cpp b/alc/effects/autowah.cpp
index 8dfee45d..581476a2 100644
--- a/alc/effects/autowah.cpp
+++ b/alc/effects/autowah.cpp
@@ -214,8 +214,8 @@ void AutowahState::process(const size_t samplesToDo,
         chandata->mFilter.z2 = z2;
 
         /* Now, mix the processed sound data to the output. */
-        MixSamples({mBufferOut, samplesToDo}, {&samplesOut[outidx], 1}, &chandata->mCurrentGain,
-            &chandata->mTargetGain, samplesToDo, 0);
+        MixSamples({mBufferOut, samplesToDo}, samplesOut[outidx].data(), chandata->mCurrentGain,
+            chandata->mTargetGain, samplesToDo);
         ++chandata;
     }
 }
diff --git a/alc/effects/equalizer.cpp b/alc/effects/equalizer.cpp
index de067911..52c491ec 100644
--- a/alc/effects/equalizer.cpp
+++ b/alc/effects/equalizer.cpp
@@ -182,8 +182,8 @@ void EqualizerState::process(const size_t samplesToDo, const al::span<const Floa
             DualBiquad{chan->mFilter[0], chan->mFilter[1]}.process(inbuf, buffer.begin());
             DualBiquad{chan->mFilter[2], chan->mFilter[3]}.process(buffer, buffer.begin());
 
-            MixSamples(buffer, {&samplesOut[outidx], 1}, &chan->mCurrentGain, &chan->mTargetGain,
-                samplesToDo, 0u);
+            MixSamples(buffer, samplesOut[outidx].data(), chan->mCurrentGain, chan->mTargetGain,
+                samplesToDo);
         }
         ++chan;
     }
diff --git a/alc/effects/modulator.cpp b/alc/effects/modulator.cpp
index 29009247..5699badf 100644
--- a/alc/effects/modulator.cpp
+++ b/alc/effects/modulator.cpp
@@ -168,8 +168,8 @@ void ModulatorState::process(const size_t samplesToDo, const al::span<const Floa
                 for(size_t i{0u};i < td;i++)
                     temps[i] *= modsamples[i];
 
-                MixSamples({temps, td}, {&samplesOut[outidx], 1}, &chandata->mCurrentGain,
-                    &chandata->mTargetGain, samplesToDo-base, base);
+                MixSamples({temps, td}, samplesOut[outidx].data()+base, chandata->mCurrentGain,
+                    chandata->mTargetGain, samplesToDo-base);
             }
             ++chandata;
         }
diff --git a/alc/effects/vmorpher.cpp b/alc/effects/vmorpher.cpp
index 869c004e..81869dca 100644
--- a/alc/effects/vmorpher.cpp
+++ b/alc/effects/vmorpher.cpp
@@ -326,8 +326,8 @@ void VmorpherState::process(const size_t samplesToDo, const al::span<const Float
                 blended[i] = lerpf(mSampleBufferA[i], mSampleBufferB[i], mLfo[i]);
 
             /* Now, mix the processed sound data to the output. */
-            MixSamples({blended, td}, {&samplesOut[outidx], 1}, &chandata->mCurrentGain,
-                &chandata->mTargetGain, samplesToDo-base, base);
+            MixSamples({blended, td}, samplesOut[outidx].data()+base, chandata->mCurrentGain,
+                chandata->mTargetGain, samplesToDo-base);
             ++chandata;
         }
 
diff --git a/core/mixer.cpp b/core/mixer.cpp
index b98a2532..066c57bd 100644
--- a/core/mixer.cpp
+++ b/core/mixer.cpp
@@ -14,6 +14,7 @@ struct CTag;
 
 
 MixerOutFunc MixSamplesOut{Mix_<CTag>};
+MixerOneFunc MixSamplesOne{Mix_<CTag>};
 
 
 std::array<float,MaxAmbiChannels> CalcAmbiCoeffs(const float y, const float z, const float x,
diff --git a/core/mixer.h b/core/mixer.h
index 9ed1f5b3..aa7597bb 100644
--- a/core/mixer.h
+++ b/core/mixer.h
@@ -24,6 +24,14 @@ inline void MixSamples(const al::span<const float> InSamples,
     const size_t Counter, const size_t OutPos)
 { MixSamplesOut(InSamples, OutBuffer, CurrentGains, TargetGains, Counter, OutPos); }
 
+/* Mixer functions that handle one input and one output channel. */
+using MixerOneFunc = void(*)(const al::span<const float> InSamples, float *OutBuffer,
+    float &CurrentGain, const float TargetGain, const size_t Counter);
+
+extern MixerOneFunc MixSamplesOne;
+inline void MixSamples(const al::span<const float> InSamples, float *OutBuffer, float &CurrentGain,
+    const float TargetGain, const size_t Counter)
+{ MixSamplesOne(InSamples, OutBuffer, CurrentGain, TargetGain, Counter); }
 
 
 /**
diff --git a/core/mixer/defs.h b/core/mixer/defs.h
index 77dae4ca..9071cb9b 100644
--- a/core/mixer/defs.h
+++ b/core/mixer/defs.h
@@ -67,6 +67,9 @@ float *Resample_(const InterpState *state, float *RESTRICT src, uint frac, uint
 template<typename InstTag>
 void Mix_(const al::span<const float> InSamples, const al::span<FloatBufferLine> OutBuffer,
     float *CurrentGains, const float *TargetGains, const size_t Counter, const size_t OutPos);
+template<typename InstTag>
+void Mix_(const al::span<const float> InSamples, float *OutBuffer, float &CurrentGain,
+    const float TargetGain, const size_t Counter);
 
 template<typename InstTag>
 void MixHrtf_(const float *InSamples, float2 *AccumSamples, const uint IrSize,
diff --git a/core/mixer/mixer_c.cpp b/core/mixer/mixer_c.cpp
index f3e6aa6a..dabfa652 100644
--- a/core/mixer/mixer_c.cpp
+++ b/core/mixer/mixer_c.cpp
@@ -198,3 +198,38 @@ void Mix_<CTag>(const al::span<const float> InSamples, const al::span<FloatBuffe
             dst[pos] += InSamples[pos] * gain;
     }
 }
+
+template<>
+void Mix_<CTag>(const al::span<const float> InSamples, float *OutBuffer, float &CurrentGain,
+    const float TargetGain, const size_t Counter)
+{
+    const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
+    const auto min_len = minz(Counter, InSamples.size());
+
+    float *RESTRICT dst{al::assume_aligned<16>(OutBuffer)};
+    float gain{CurrentGain};
+    const float step{(TargetGain-gain) * delta};
+
+    size_t pos{0};
+    if(!(std::abs(step) > std::numeric_limits<float>::epsilon()))
+        gain = TargetGain;
+    else
+    {
+        float step_count{0.0f};
+        for(;pos != min_len;++pos)
+        {
+            dst[pos] += InSamples[pos] * (gain + step*step_count);
+            step_count += 1.0f;
+        }
+        if(pos == Counter)
+            gain = TargetGain;
+        else
+            gain += step*step_count;
+    }
+    CurrentGain = gain;
+
+    if(!(std::abs(gain) > GainSilenceThreshold))
+        return;
+    for(;pos != InSamples.size();++pos)
+        dst[pos] += InSamples[pos] * gain;
+}
diff --git a/core/mixer/mixer_neon.cpp b/core/mixer/mixer_neon.cpp
index a3468926..f3d54fec 100644
--- a/core/mixer/mixer_neon.cpp
+++ b/core/mixer/mixer_neon.cpp
@@ -305,3 +305,80 @@ void Mix_<NEONTag>(const al::span<const float> InSamples, const al::span<FloatBu
             dst[pos] += InSamples[pos] * gain;
     }
 }
+
+template<>
+void Mix_<NEONTag>(const al::span<const float> InSamples, float *OutBuffer, float &CurrentGain,
+    const float TargetGain, const size_t Counter)
+{
+    const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
+    const auto min_len = minz(Counter, InSamples.size());
+    const auto aligned_len = minz((min_len+3) & ~size_t{3}, InSamples.size()) - min_len;
+
+    float *RESTRICT dst{al::assume_aligned<16>(OutBuffer)};
+    float gain{CurrentGain};
+    const float step{(TargetGain-gain) * delta};
+
+    size_t pos{0};
+    if(!(std::abs(step) > std::numeric_limits<float>::epsilon()))
+        gain = TargetGain;
+    else
+    {
+        float step_count{0.0f};
+        /* Mix with applying gain steps in aligned multiples of 4. */
+        if(size_t todo{min_len >> 2})
+        {
+            const float32x4_t four4{vdupq_n_f32(4.0f)};
+            const float32x4_t step4{vdupq_n_f32(step)};
+            const float32x4_t gain4{vdupq_n_f32(gain)};
+            float32x4_t step_count4{vdupq_n_f32(0.0f)};
+            step_count4 = vsetq_lane_f32(1.0f, step_count4, 1);
+            step_count4 = vsetq_lane_f32(2.0f, step_count4, 2);
+            step_count4 = vsetq_lane_f32(3.0f, step_count4, 3);
+
+            do {
+                const float32x4_t val4 = vld1q_f32(&InSamples[pos]);
+                float32x4_t dry4 = vld1q_f32(&dst[pos]);
+                dry4 = vmlaq_f32(dry4, val4, vmlaq_f32(gain4, step4, step_count4));
+                step_count4 = vaddq_f32(step_count4, four4);
+                vst1q_f32(&dst[pos], dry4);
+                pos += 4;
+            } while(--todo);
+            /* NOTE: step_count4 now represents the next four counts after the
+             * last four mixed samples, so the lowest element represents the
+             * next step count to apply.
+             */
+            step_count = vgetq_lane_f32(step_count4, 0);
+        }
+        /* Mix with applying left over gain steps that aren't aligned multiples of 4. */
+        for(size_t leftover{min_len&3};leftover;++pos,--leftover)
+        {
+            dst[pos] += InSamples[pos] * (gain + step*step_count);
+            step_count += 1.0f;
+        }
+        if(pos == Counter)
+            gain = TargetGain;
+        else
+            gain += step*step_count;
+
+        /* Mix until pos is aligned with 4 or the mix is done. */
+        for(size_t leftover{aligned_len&3};leftover;++pos,--leftover)
+            dst[pos] += InSamples[pos] * gain;
+    }
+    CurrentGain = gain;
+
+    if(!(std::abs(gain) > GainSilenceThreshold))
+        return;
+    if(size_t todo{(InSamples.size()-pos) >> 2})
+    {
+        const float32x4_t gain4 = vdupq_n_f32(gain);
+        do {
+            const float32x4_t val4 = vld1q_f32(&InSamples[pos]);
+            float32x4_t dry4 = vld1q_f32(&dst[pos]);
+            dry4 = vmlaq_f32(dry4, val4, gain4);
+            vst1q_f32(&dst[pos], dry4);
+            pos += 4;
+        } while(--todo);
+    }
+    for(size_t leftover{(InSamples.size()-pos)&3};leftover;++pos,--leftover)
+        dst[pos] += InSamples[pos] * gain;
+}
diff --git a/core/mixer/mixer_sse.cpp b/core/mixer/mixer_sse.cpp
index fc4d8cc6..89968fc8 100644
--- a/core/mixer/mixer_sse.cpp
+++ b/core/mixer/mixer_sse.cpp
@@ -270,3 +270,79 @@ void Mix_<SSETag>(const al::span<const float> InSamples, const al::span<FloatBuf
             dst[pos] += InSamples[pos] * gain;
     }
 }
+
+template<>
+void Mix_<SSETag>(const al::span<const float> InSamples, float *OutBuffer, float &CurrentGain,
+    const float TargetGain, const size_t Counter)
+{
+    const float delta{(Counter > 0) ? 1.0f / static_cast<float>(Counter) : 0.0f};
+    const auto min_len = minz(Counter, InSamples.size());
+    const auto aligned_len = minz((min_len+3) & ~size_t{3}, InSamples.size()) - min_len;
+
+    float *RESTRICT dst{al::assume_aligned<16>(OutBuffer)};
+    float gain{CurrentGain};
+    const float step{(TargetGain-gain) * delta};
+
+    size_t pos{0};
+    if(!(std::abs(step) > std::numeric_limits<float>::epsilon()))
+        gain = TargetGain;
+    else
+    {
+        float step_count{0.0f};
+        /* Mix with applying gain steps in aligned multiples of 4. */
+        if(size_t todo{min_len >> 2})
+        {
+            const __m128 four4{_mm_set1_ps(4.0f)};
+            const __m128 step4{_mm_set1_ps(step)};
+            const __m128 gain4{_mm_set1_ps(gain)};
+            __m128 step_count4{_mm_setr_ps(0.0f, 1.0f, 2.0f, 3.0f)};
+            do {
+                const __m128 val4{_mm_load_ps(&InSamples[pos])};
+                __m128 dry4{_mm_load_ps(&dst[pos])};
+
+                /* dry += val * (gain + step*step_count) */
+                dry4 = MLA4(dry4, val4, MLA4(gain4, step4, step_count4));
+
+                _mm_store_ps(&dst[pos], dry4);
+                step_count4 = _mm_add_ps(step_count4, four4);
+                pos += 4;
+            } while(--todo);
+            /* NOTE: step_count4 now represents the next four counts after the
+             * last four mixed samples, so the lowest element represents the
+             * next step count to apply.
+             */
+            step_count = _mm_cvtss_f32(step_count4);
+        }
+        /* Mix with applying left over gain steps that aren't aligned multiples of 4. */
+        for(size_t leftover{min_len&3};leftover;++pos,--leftover)
+        {
+            dst[pos] += InSamples[pos] * (gain + step*step_count);
+            step_count += 1.0f;
+        }
+        if(pos == Counter)
+            gain = TargetGain;
+        else
+            gain += step*step_count;
+
+        /* Mix until pos is aligned with 4 or the mix is done. */
+        for(size_t leftover{aligned_len&3};leftover;++pos,--leftover)
+            dst[pos] += InSamples[pos] * gain;
+    }
+    CurrentGain = gain;
+
+    if(!(std::abs(gain) > GainSilenceThreshold))
+        return;
+    if(size_t todo{(InSamples.size()-pos) >> 2})
+    {
+        const __m128 gain4{_mm_set1_ps(gain)};
+        do {
+            const __m128 val4{_mm_load_ps(&InSamples[pos])};
+            __m128 dry4{_mm_load_ps(&dst[pos])};
+            dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
+            _mm_store_ps(&dst[pos], dry4);
+            pos += 4;
+        } while(--todo);
+    }
+    for(size_t leftover{(InSamples.size()-pos)&3};leftover;++pos,--leftover)
+        dst[pos] += InSamples[pos] * gain;
+}
diff --git a/core/voice.cpp b/core/voice.cpp
index 1b979614..5fa9f677 100644
--- a/core/voice.cpp
+++ b/core/voice.cpp
@@ -84,6 +84,19 @@ inline MixerOutFunc SelectMixer()
     return Mix_<CTag>;
 }
 
+inline MixerOneFunc SelectMixerOne()
+{
+#ifdef HAVE_NEON
+    if((CPUCapFlags&CPU_CAP_NEON))
+        return Mix_<NEONTag>;
+#endif
+#ifdef HAVE_SSE
+    if((CPUCapFlags&CPU_CAP_SSE))
+        return Mix_<SSETag>;
+#endif
+    return Mix_<CTag>;
+}
+
 inline HrtfMixerFunc SelectHrtfMixer()
 {
 #ifdef HAVE_NEON
@@ -153,6 +166,7 @@ void Voice::InitMixer(al::optional<std::string> resampler)
     }
 
     MixSamplesOut = SelectMixer();
+    MixSamplesOne = SelectMixerOne();
     MixHrtfBlendSamples = SelectHrtfBlendMixer();
     MixHrtfSamples = SelectHrtfMixer();
 }