HalfFloat_SSE2 port from C algorithm to SSE2

source/planar_functions.cc

Index: source/planar_functions.cc
diff --git a/source/planar_functions.cc b/source/planar_functions.cc
index 20e9c66c0b88fab568318884e89784c32adc67a9..67e01f45cb0e36bb7952de90cf505d120004fe21 100644
--- a/source/planar_functions.cc
+++ b/source/planar_functions.cc
@@ -2470,6 +2470,14 @@ int HalfFloatPlane(const uint16* src_y, int src_stride_y,
     height = 1;
     src_stride_y = dst_stride_y = 0;
   }
+#if defined(HAS_HALFFLOATROW_SSE2)
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    HalfFloatRow = HalfFloatRow_Any_SSE2;
+    if (IS_ALIGNED(width, 8)) {
+      HalfFloatRow = HalfFloatRow_SSE2;
+    }
+  }
+#endif
 #if defined(HAS_HALFFLOATROW_AVX2)
   if (TestCpuFlag(kCpuHasAVX2)) {
     HalfFloatRow = HalfFloatRow_Any_AVX2;
@@ -2478,6 +2486,14 @@ int HalfFloatPlane(const uint16* src_y, int src_stride_y,
     }
   }
 #endif
+#if defined(HAS_HALFFLOATROW_AVX)
+  if (TestCpuFlag(kCpuHasAVX)) {
+//    HalfFloatRow = HalfFloatRow_Any_AVX2;
+    if (IS_ALIGNED(width, 16)) {
+      HalfFloatRow = HalfFloatRow_AVX;
+    }
+  }
+#endif
 
   for (y = 0; y < height; ++y) {
     HalfFloatRow(src_y, dst_y, scale, width);