From b7da69510c85b776ba119d69c4c74aa38d412594 Mon Sep 17 00:00:00 2001
From: Chris Robinson <chris.kcat@gmail.com>
Date: Wed, 1 Jun 2016 23:39:13 -0700
Subject: [PATCH] Implement a Neon-enhanced MixRow

---
 Alc/bformatdec.c |  4 ++++
 Alc/mixer_defs.h |  2 ++
 Alc/mixer_neon.c | 25 +++++++++++++++++++++++++
 3 files changed, 31 insertions(+)

diff --git a/Alc/bformatdec.c b/Alc/bformatdec.c
index 7da50692..9ebaba27 100644
--- a/Alc/bformatdec.c
+++ b/Alc/bformatdec.c
@@ -159,6 +159,10 @@ static inline MatrixMixerFunc SelectMixer(void)
     if((CPUCapFlags&CPU_CAP_SSE))
         return MixRow_SSE;
 #endif
+#ifdef HAVE_NEON
+    if((CPUCapFlags&CPU_CAP_NEON))
+        return MixRow_Neon;
+#endif
     return MixRow_C;
 }
 
diff --git a/Alc/mixer_defs.h b/Alc/mixer_defs.h
index 8d208b9c..8b934c58 100644
--- a/Alc/mixer_defs.h
+++ b/Alc/mixer_defs.h
@@ -80,5 +80,7 @@ void MixHrtf_Neon(ALfloat (*restrict OutBuffer)[BUFFERSIZE], ALuint lidx, ALuint
                   struct HrtfState *hrtfstate, ALuint BufferSize);
 void Mix_Neon(const ALfloat *data, ALuint OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
               struct MixGains *Gains, ALuint Counter, ALuint OutPos, ALuint BufferSize);
+void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Mtx, ALfloat (*restrict data)[BUFFERSIZE],
+                 ALuint InChans, ALuint BufferSize);
 
 #endif /* MIXER_DEFS_H */
diff --git a/Alc/mixer_neon.c b/Alc/mixer_neon.c
index 96936ef5..073f62c8 100644
--- a/Alc/mixer_neon.c
+++ b/Alc/mixer_neon.c
@@ -118,3 +118,28 @@ void Mix_Neon(const ALfloat *data, ALuint OutChans, ALfloat (*restrict OutBuffer
             OutBuffer[c][OutPos+pos] += data[pos]*gain;
     }
 }
+
+void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Mtx, ALfloat (*restrict data)[BUFFERSIZE], ALuint InChans, ALuint BufferSize)
+{
+    float32x4_t gain4;
+    ALuint c;
+
+    for(c = 0;c < InChans;c++)
+    {
+        ALuint pos = 0;
+        ALfloat gain = Mtx[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][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][pos]*gain;
+    }
+}
-- 
2.11.4.GIT