Raw 24 bit RGB to RGB24 (bgr)

source/row_neon.cc

 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 780 matching lines @@
   "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)
     "vst4.8     {d1, d2, d3, d4}, [%1]!        \n"  // store 8 pixels of ARGB.
     "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 */ \
     "vshl.u8    d6, d6, #2                     \n"  /* G GGGGGG00 upper 6   */ \
     "vshr.u8    d1, d1, #3                     \n"  /* R 000RRRRR lower 5   */ \
     "vshl.u8    q0, q0, #3                     \n"  /* B,R BBBBB000 upper 5 */ \
     "vshr.u8    q2, q0, #5                     \n"  /* B,R 00000BBB lower 3 */ \
     "vorr.u8    d0, d0, d4                     \n"  /* B                    */ \
     "vshr.u8    d4, d6, #6                     \n"  /* G 000000GG lower 2   */ \
     "vorr.u8    d2, d1, d5                     \n"  /* R                    */ \
@@ skipping 2042 matching lines @@
     "r"(6)             // %5
   : "cc", "memory", "q0", "q1"  // Clobber List
   );
 }
 #endif  // defined(__ARM_NEON__) && !defined(__aarch64__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif