ARGBShuffle_AVX2 for speed up end swapping for Chrome/Java.

BUG=271
TESTED=ARGBShuffle unittest
R=mflodman@webrtc.org

Review URL: https://webrtc-codereview.appspot.com/2320005

git-svn-id: http://libyuv.googlecode.com/svn/trunk@804 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 803
+Version: 804
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -150,11 +150,14 @@ extern "C" {
 #define GCC_HAS_AVX2 1
 #endif  // GNUC >= 4.7
 #endif  // __GNUC__
+// TODO(fbarchard): Test with new NaCL tool chain.  Change __native_client__AVX2
+// to __native_client__ to test.
 #if !defined(LIBYUV_DISABLE_X86) && \
   ((defined(_M_IX86) && defined(_MSC_VER) && _MSC_VER >= 1700) || \
-  defined(__native_client__) || defined(__clang__) || defined(GCC_HAS_AVX2))
+  defined(__native_client__AVX2) || defined(__clang__) || defined(GCC_HAS_AVX2))
 // Effects:
 #define HAS_ARGBPOLYNOMIALROW_AVX2
+#define HAS_ARGBSHUFFLEROW_AVX2
 #endif
 
 // The following are Windows only:
@@ -166,7 +169,6 @@ extern "C" {
 
 // Caveat: Visual C 2012 required for AVX2.
 #if _MSC_VER >= 1700
-#define HAS_ARGBSHUFFLEROW_AVX2
 #define HAS_ARGBTOUVROW_AVX2
 #define HAS_ARGBTOYJROW_AVX2
 #define HAS_ARGBTOYROW_AVX2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 803
+#define LIBYUV_VERSION 804
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/planar_functions.cc

@@ -1592,7 +1592,7 @@ int ARGBBlur(const uint8* src_argb, int src_stride_argb,
     int x;
     for (x = 0; x < radius + 1; ++x) {
       CumulativeSumToAverageRow(cumsum_top_row, cumsum_bot_row,
-                              boxwidth, area, &dst_argb[x * 4], 1);
+                                boxwidth, area, &dst_argb[x * 4], 1);
       area += (bot_y - top_y);
       boxwidth += 4;
     }
@@ -1600,15 +1600,15 @@ int ARGBBlur(const uint8* src_argb, int src_stride_argb,
     // Middle unclipped.
     int n = (width - 1) - radius - x + 1;
     CumulativeSumToAverageRow(cumsum_top_row, cumsum_bot_row,
-                           boxwidth, area, &dst_argb[x * 4], n);
+                              boxwidth, area, &dst_argb[x * 4], n);
 
     // Right clipped.
     for (x += n; x <= width - 1; ++x) {
       area -= (bot_y - top_y);
       boxwidth -= 4;
       CumulativeSumToAverageRow(cumsum_top_row + (x - radius - 1) * 4,
-                             cumsum_bot_row + (x - radius - 1) * 4,
-                             boxwidth, area, &dst_argb[x * 4], 1);
+                                cumsum_bot_row + (x - radius - 1) * 4,
+                                boxwidth, area, &dst_argb[x * 4], 1);
     }
     dst_argb += dst_stride_argb;
   }

source/row_posix.cc

@@ -5726,6 +5726,38 @@ void ARGBShuffleRow_Unaligned_SSSE3(const uint8* src_argb, uint8* dst_argb,
 }
 #endif  // HAS_ARGBSHUFFLEROW_SSSE3
 
+#ifdef HAS_ARGBSHUFFLEROW_AVX2
+// For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
+void ARGBShuffleRow_AVX2(const uint8* src_argb, uint8* dst_argb,
+                         const uint8* shuffler, int pix) {
+  asm volatile (
+    "vmovdqa   "MEMACCESS(3)",%%xmm5           \n"
+    "vpermq    $0x44,%%ymm5,%%ymm5             \n"
+
+    ".p2align  4                               \n"
+  "1:                                          \n"
+    "vmovdqu   "MEMACCESS(0)",%%ymm0           \n"
+    "vmovdqu   "MEMACCESS2(0x20,0)",%%ymm1     \n"
+    "lea       "MEMLEA(0x40,0)",%0             \n"
+    "vpshufb   %%ymm5,%%ymm0,%%ymm0            \n"
+    "vpshufb   %%ymm5,%%ymm1,%%ymm1            \n"
+    "sub       $0x10,%2                        \n"
+    "vmovdqu   %%ymm0,"MEMACCESS(1)"           \n"
+    "vmovdqu   %%ymm1,"MEMACCESS2(0x20,1)"     \n"
+    "lea       "MEMLEA(0x40,1)",%1             \n"
+    "jg        1b                              \n"
+  : "+r"(src_argb),  // %0
+    "+r"(dst_argb),  // %1
+    "+r"(pix)        // %2
+  : "r"(shuffler)    // %3
+  : "memory", "cc"
+#if defined(__SSE2__)
+    , "xmm0", "xmm1", "xmm5"
+#endif
+  );
+}
+#endif  // HAS_ARGBSHUFFLEROW_AVX2
+
 #ifdef HAS_I422TOYUY2ROW_SSE2
 void I422ToYUY2Row_SSE2(const uint8* src_y,
                         const uint8* src_u,