Add Sk4f::ToBytes(uint8_t[16], Sk4f, Sk4f, Sk4f, Sk4f)

src/opts/SkNx_sse.h

Index: src/opts/SkNx_sse.h
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index f0ccd3f7b59128e9e5b7d23468e529ee3e687a68..a4f8656536c4bf3facdb7c5df769d81770f3800c 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -132,6 +132,15 @@ public:
         *(int*)bytes = _mm_cvtsi128_si32(fix8);
     }
 
+    static void ToBytes(uint8_t bytes[16],
+                        const SkNx& a, const SkNx& b, const SkNx& c, const SkNx& d) {
+        _mm_storeu_si128((__m128i*)bytes,
+                         _mm_packus_epi16(_mm_packus_epi16(_mm_cvttps_epi32(a.fVec),
+                                                           _mm_cvttps_epi32(b.fVec)),
+                                          _mm_packus_epi16(_mm_cvttps_epi32(c.fVec),
+                                                           _mm_cvttps_epi32(d.fVec))));
+    }
+
     SkNx operator + (const SkNx& o) const { return _mm_add_ps(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_ps(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return _mm_mul_ps(fVec, o.fVec); }