ARGBAdd ported AVX2 ported to GCC.

BUG=269
TESTED=try bots
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/32449004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1158 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1157
+Version: 1158
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -185,7 +185,6 @@ extern "C" {
 // The code supports NaCL but requires a new compiler and validator.
 #if !defined(LIBYUV_DISABLE_X86) && (defined(VISUALC_HAS_AVX2) || \
     defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2))
-// Effects:
 #define HAS_COPYROW_AVX
 #define HAS_ARGBPOLYNOMIALROW_AVX2
 #define HAS_ARGBSHUFFLEROW_AVX2
@@ -200,6 +199,9 @@ extern "C" {
 #define HAS_UYVYTOUVROW_AVX2
 #define HAS_SPLITUVROW_AVX2
 #define HAS_MERGEUVROW_AVX2
+
+// Effects:
+#define HAS_ARGBADDROW_AVX2
 #endif
 
 // The following are require VS2012.
@@ -215,7 +217,6 @@ extern "C" {
 #define HAS_MIRRORROW_AVX2
 
 // Effects:
-#define HAS_ARGBADDROW_AVX2
 #define HAS_ARGBATTENUATEROW_AVX2
 #define HAS_ARGBMIRRORROW_AVX2
 #define HAS_ARGBMULTIPLYROW_AVX2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1157
+#define LIBYUV_VERSION 1158
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_posix.cc

@@ -3955,6 +3955,36 @@ void ARGBAddRow_SSE2(const uint8* src_argb0, const uint8* src_argb1,
 }
 #endif  // HAS_ARGBADDROW_SSE2
 
+#ifdef HAS_ARGBADDROW_AVX2
+// Add 2 rows of ARGB pixels together, 4 pixels at a time.
+void ARGBAddRow_AVX2(const uint8* src_argb0, const uint8* src_argb1,
+                     uint8* dst_argb, int width) {
+  asm volatile (
+    // 4 pixel loop.
+    LABELALIGN
+  "1:                                          \n"
+    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
+    "lea        " MEMLEA(0x20,0) ",%0          \n"
+    "vpaddusb   " MEMACCESS(1) ",%%ymm0,%%ymm0 \n"
+    "lea        " MEMLEA(0x20,1) ",%1          \n"
+    "vmovdqu    %%ymm0," MEMACCESS(2) "        \n"
+    "lea        " MEMLEA(0x20,2) ",%2          \n"
+    "sub        $0x8,%3                        \n"
+    "jg        1b                              \n"
+    "vzeroupper                                \n"
+  : "+r"(src_argb0),  // %0
+    "+r"(src_argb1),  // %1
+    "+r"(dst_argb),   // %2
+    "+r"(width)       // %3
+  :
+  : "memory", "cc"
+#if defined(__SSE2__)
+    , "xmm0"
+#endif
+  );
+}
+#endif  // HAS_ARGBADDROW_AVX2
+
 #ifdef HAS_ARGBSUBTRACTROW_SSE2
 // Subtract 2 rows of ARGB pixels, 4 pixels at a time.
 void ARGBSubtractRow_SSE2(const uint8* src_argb0, const uint8* src_argb1,