Port HalfFloatRow_SSE2 to AVX2 but not using F16C.

source/planar_functions.cc

Index: source/planar_functions.cc
diff --git a/source/planar_functions.cc b/source/planar_functions.cc
index 0bd76bc7cf9a8490dbeb9c579857f7a565b8c1fb..da4c47b831bc8e3b0a79d557adc9dbc0b5491f7c 100644
--- a/source/planar_functions.cc
+++ b/source/planar_functions.cc
@@ -2570,13 +2570,21 @@ int HalfFloatPlane(const uint16* src_y, int src_stride_y,
   }
 #endif
 #if defined(HAS_HALFFLOATROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2) && TestCpuFlag(kCpuHasF16C)) {
+  if (TestCpuFlag(kCpuHasAVX2)) {
     HalfFloatRow = HalfFloatRow_Any_AVX2;
     if (IS_ALIGNED(width, 16)) {
       HalfFloatRow = HalfFloatRow_AVX2;
     }
   }
 #endif
+#if defined(HAS_HALFFLOATROW_F16C)
+  if (TestCpuFlag(kCpuHasAVX2) && TestCpuFlag(kCpuHasF16C)) {
+    HalfFloatRow = HalfFloatRow_Any_F16C;
+    if (IS_ALIGNED(width, 16)) {
+      HalfFloatRow = HalfFloatRow_F16C;
+    }
+  }
+#endif
   for (y = 0; y < height; ++y) {
     HalfFloatRow(src_y, dst_y, scale, width);
     src_y += src_stride_y;