Raw 24 bit RGB to RGB24 (bgr)

source/row_neon.cc

Index: source/row_neon.cc
diff --git a/source/row_neon.cc b/source/row_neon.cc
index b7a3be01d54cf423005ef75190f838933658744f..24457d53d19ac945528c58c9d8871dcf0e5b11e7 100644
--- a/source/row_neon.cc
+++ b/source/row_neon.cc
@@ -798,12 +798,30 @@ void RAWToARGBRow_NEON(const uint8* src_raw, uint8* dst_argb, int width) {
     "bgt        1b                             \n"
   : "+r"(src_raw),   // %0
     "+r"(dst_argb),  // %1
-    "+r"(width)        // %2
+    "+r"(width)      // %2
   :
   : "cc", "memory", "d1", "d2", "d3", "d4"  // Clobber List
   );
 }
 
+void RAWToRGB24Row_NEON(const uint8* src_raw, uint8* dst_rgb24, int width) {
+  asm volatile (
+  "1:                                          \n"
+    MEMACCESS(0)
+    "vld3.8     {d1, d2, d3}, [%0]!            \n"  // load 8 pixels of RAW.
+    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "vswp.u8    d1, d3                         \n"  // swap R, B
+    MEMACCESS(1)
+    "vst3.8     {d1, d2, d3}, [%1]!            \n"  // store 8 pixels of RGB24.
+    "bgt        1b                             \n"
+  : "+r"(src_raw),    // %0
+    "+r"(dst_rgb24),  // %1
+    "+r"(width)       // %2
+  :
+  : "cc", "memory", "d1", "d2", "d3"  // Clobber List
+  );
+}
+
 #define RGB565TOARGB                                                           \
     "vshrn.u16  d6, q0, #5                     \n"  /* G xxGGGGGG           */ \
     "vuzp.u8    d0, d1                         \n"  /* d0 xxxBBBBB RRRRRxxx */ \