test for unaligned vs aligned for CopyRow_SSE2

source/row_gcc.cc

 // VERSION 2
 /*
  *  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 2708 matching lines @@
   :
   : "memory", "cc", NACL_R14
     "xmm0", "xmm1", "xmm2"
   );
 }
 #endif  // HAS_MERGEUVROW_SSE2
 
 #ifdef HAS_COPYROW_SSE2
 void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
   asm volatile (
+    "test       $0xf,%0                        \n"
+    "jne        2f                             \n"
+    "test       $0xf,%1                        \n"
+    "jne        2f                             \n"
     LABELALIGN
   "1:                                          \n"
+    "movdqa    " MEMACCESS(0) ",%%xmm0         \n"
+    "movdqa    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
+    "lea       " MEMLEA(0x20,0) ",%0           \n"
+    "movdqa    %%xmm0," MEMACCESS(1) "         \n"
+    "movdqa    %%xmm1," MEMACCESS2(0x10,1) "   \n"
+    "lea       " MEMLEA(0x20,1) ",%1           \n"
+    "sub       $0x20,%2                        \n"
+    "jg        1b                              \n"
+    "jmp       9f                              \n"
+    LABELALIGN
+  "2:                                          \n"
     "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
     "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
     "lea       " MEMLEA(0x20,0) ",%0           \n"
     "movdqu    %%xmm0," MEMACCESS(1) "         \n"
     "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
     "lea       " MEMLEA(0x20,1) ",%1           \n"
     "sub       $0x20,%2                        \n"
-    "jg        1b                              \n"
+    "jg        2b                              \n"
+  "9:                                          \n"
   : "+r"(src),   // %0
     "+r"(dst),   // %1
     "+r"(count)  // %2
   :
   : "memory", "cc"
     , "xmm0", "xmm1"
   );
 }
 #endif  // HAS_COPYROW_SSE2
 
@@ skipping 2713 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif