From 1b92fe75a14b6a11a04695eb52a651f5d958450a Mon Sep 17 00:00:00 2001
From: James Zern <jzern@google.com>
Date: Thu, 5 Dec 2019 21:00:45 -0800
Subject: [PATCH] DC16_NEON,aarch64: use vaddlv

saves 3 instructions, neutral to mildly faster on a pixel 3a

Change-Id: I6ae57e8e38d4149167ea14e27cd2b32113b4f8e7
---
 src/dsp/dec_neon.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/dsp/dec_neon.c b/src/dsp/dec_neon.c
index 0b99f00a..37e2e69d 100644
--- a/src/dsp/dec_neon.c
+++ b/src/dsp/dec_neon.c
@@ -1511,11 +1511,16 @@ static WEBP_INLINE void DC16_NEON(uint8_t* dst, int do_top, int do_left) {
 
   if (do_top) {
     const uint8x16_t A = vld1q_u8(dst - BPS);  // top row
+#if defined(__aarch64__)
+    const uint16_t p3 = vaddlvq_u8(A);
+    sum_top = vdupq_n_u16(p3);
+#else
     const uint16x8_t p0 = vpaddlq_u8(A);  // cascading summation of the top
     const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
     const uint16x4_t p2 = vpadd_u16(p1, p1);
     const uint16x4_t p3 = vpadd_u16(p2, p2);
     sum_top = vcombine_u16(p3, p3);
+#endif
   }
 
   if (do_left) {