Refactor to put SkXfermode_opts inside SK_OPTS_NS.

src/opts/SkPMFloat_sse.h

Index: src/opts/SkPMFloat_sse.h
diff --git a/src/opts/SkPMFloat_sse.h b/src/opts/SkPMFloat_sse.h
index 28aa90bf29fd8d4913417f8111d92f1bc9b23448..016644e0524255ed7a65918cf2753ad7ed28bb7d 100644
--- a/src/opts/SkPMFloat_sse.h
+++ b/src/opts/SkPMFloat_sse.h
@@ -5,9 +5,7 @@
  * found in the LICENSE file.
  */
 
-namespace {  // See SkPMFloat.h
-
-inline SkPMFloat::SkPMFloat(SkPMColor c) {
+SK_ALWAYS_INLINE SkPMFloat::SkPMFloat(SkPMColor c) {
     SkPMColorAssert(c);
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
     const int _ = 255;  // Zero these bytes.
@@ -22,7 +20,7 @@ inline SkPMFloat::SkPMFloat(SkPMColor c) {
     SkASSERT(this->isValid());
 }
 
-inline SkPMColor SkPMFloat::round() const {
+SK_ALWAYS_INLINE SkPMColor SkPMFloat::round() const {
     // We don't use _mm_cvtps_epi32, because we want precise control over how 0.5 rounds (up).
     __m128 scaled = _mm_mul_ps(_mm_set1_ps(255), fVec);
     __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), scaled)),
@@ -33,9 +31,7 @@ inline SkPMColor SkPMFloat::round() const {
     return c;
 }
 
-inline Sk4f SkPMFloat::alphas() const {
+SK_ALWAYS_INLINE Sk4f SkPMFloat::alphas() const {
     static_assert(SK_A32_SHIFT == 24, "");
     return _mm_shuffle_ps(fVec, fVec, 0xff);  // Read as 11 11 11 11, copying lane 3 to all lanes.
 }
-
-}  // namespace