From 92753e8b8a384df005af778db456683fea1481f7 Mon Sep 17 00:00:00 2001
From: Themaister <maister@archlinux.us>
Date: Sun, 29 Dec 2013 12:59:57 +0100
Subject: [PATCH] Add volume control to NEON ASM.

---
 audio/utils.c      | 8 +++-----
 audio/utils_neon.S | 6 +++++-
 2 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/audio/utils.c b/audio/utils.c
index 3b4e7963b1..09f18baa74 100644
--- a/audio/utils.c
+++ b/audio/utils.c
@@ -140,19 +140,17 @@ void audio_convert_float_to_s16_altivec(int16_t *out,
       audio_convert_float_to_s16_C(out, in, samples);
 }
 #elif defined(HAVE_NEON)
-void audio_convert_s16_float_asm(float *out, const int16_t *in, size_t samples);
+void audio_convert_s16_float_asm(float *out, const int16_t *in, size_t samples, const float *gain); // Avoid potential hard-float/soft-float ABI issues.
 static void audio_convert_s16_to_float_neon(float *out, const int16_t *in, size_t samples,
       float gain)
 {
-   (void)gain; // gain is ignored for now.
-
    size_t aligned_samples = samples & ~7;
    if (aligned_samples)
-      audio_convert_s16_float_asm(out, in, aligned_samples);
+      audio_convert_s16_float_asm(out, in, aligned_samples, &gain);
 
    // Could do all conversion in ASM, but keep it simple for now.
    audio_convert_s16_to_float_C(out + aligned_samples, in + aligned_samples,
-         samples - aligned_samples, 1.0f);
+         samples - aligned_samples, gain);
 }
 
 void audio_convert_float_s16_asm(int16_t *out, const float *in, size_t samples);
diff --git a/audio/utils_neon.S b/audio/utils_neon.S
index cf433fd90e..c0fc53ed75 100644
--- a/audio/utils_neon.S
+++ b/audio/utils_neon.S
@@ -17,7 +17,7 @@
 
 .align 4
 .globl audio_convert_s16_float_asm
-# audio_convert_s16_float_asm(float *out, const int16_t *in, size_t samples)
+# audio_convert_s16_float_asm(float *out, const int16_t *in, size_t samples, const float *gain)
 audio_convert_s16_float_asm:
    # Hacky way to get a constant of 2^-15.
    # Might be faster to just load a constant from memory.
@@ -28,6 +28,10 @@ audio_convert_s16_float_asm:
    vmul.f32 q8, q8, q8
    vadd.f32 q8, q8, q8
 
+   # Apply gain
+   vld1.f32 {d6[0]}, [r3]
+   vmul.f32 q8, q8, d6[0]
+
 1:
    # Preload here?
    vld1.s16 {q0}, [r1]!