ARGBExtractAlpha 16 pixels at a time for ARM

source/row_neon64.cc

Index: source/row_neon64.cc
diff --git a/source/row_neon64.cc b/source/row_neon64.cc
index e5f2dc8f396a17a93c0c890c875770fbdec677a0..608241691a4c8a08289629cdb484590d88ab9975 100644
--- a/source/row_neon64.cc
+++ b/source/row_neon64.cc
@@ -1404,10 +1404,10 @@ void ARGBExtractAlphaRow_NEON(const uint8* src_argb, uint8* dst_a, int width) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
-    "ld4        {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"  // load row 8 pixels
-    "subs       %w2, %w2, #8                   \n"  // 8 processed per loop
+    "ld4        {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n"  // load row 16 pixels
+    "subs       %w2, %w2, #16                  \n"  // 16 processed per loop
     MEMACCESS(1)
-    "st1        {v3.8b}, [%1], #8              \n"  // store 8 A's.
+    "st1        {v3.16b}, [%1], #16            \n"  // store 16 A's.
     "b.gt       1b                             \n"
   : "+r"(src_argb),   // %0
     "+r"(dst_a),      // %1