ARGBExtractAlpha 16 pixels at a time for ARM

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 1284 matching lines @@
     "+r"(width)        // %2
   :
   : "cc", "memory", "q0", "q1", "q2", "q12", "q13"
   );
 }
 
 void ARGBExtractAlphaRow_NEON(const uint8* src_argb, uint8* dst_a, int width) {
   asm volatile (
   "1:                                          \n"
     MEMACCESS(0)
-    "vld4.8     {d0, d1, d2, d3}, [%0]!        \n"  // load row 8 pixels
-    "subs       %2, %2, #8                     \n"  // 8 processed per loop
+    "vld4.8     {d0, d2, d4, d6}, [%0]!        \n"  // load 8 ARGB pixels
+    "vld4.8     {d1, d3, d5, d7}, [%0]!        \n"  // load next 8 ARGB pixels
+    "subs       %2, %2, #16                    \n"  // 16 processed per loop
     MEMACCESS(1)
-    "vst1.8     {d3}, [%1]!                    \n"  // store 8 A's.
+    "vst1.8     {q3}, [%1]!                    \n"  // store 16 A's.
     "bgt       1b                              \n"
   : "+r"(src_argb),   // %0
     "+r"(dst_a),      // %1
     "+r"(width)       // %2
   :
-  : "cc", "memory", "d0", "d1", "d2", "d3"  // Clobber List
+  : "cc", "memory", "q0", "q1", "q2", "q3"  // Clobber List
   );
 }
 
 void ARGBToYJRow_NEON(const uint8* src_argb, uint8* dst_y, int width) {
   asm volatile (
     "vmov.u8    d24, #15                       \n"  // B * 0.11400 coefficient
     "vmov.u8    d25, #75                       \n"  // G * 0.58700 coefficient
     "vmov.u8    d26, #38                       \n"  // R * 0.29900 coefficient
   "1:                                          \n"
     MEMACCESS(0)
@@ skipping 1511 matching lines @@
     "r"(6)             // %5
   : "cc", "memory", "q0", "q1"  // Clobber List
   );
 }
 #endif  // defined(__ARM_NEON__) && !defined(__aarch64__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif