Specialize Sk2d for SSE2

src/opts/SkNx_sse.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkNx_sse_DEFINED
 #define SkNx_sse_DEFINED
 
@@ skipping 47 matching lines @@
         return pun.fs[k&1];
     }
 
     bool allTrue() const { return 0xff == (_mm_movemask_epi8(_mm_castps_si128(fVec)) & 0xff); }
     bool anyTrue() const { return 0x00 != (_mm_movemask_epi8(_mm_castps_si128(fVec)) & 0xff); }
 
     __m128 fVec;
 };
 
 template <>
+class SkNx<2, double> {
+public:
+    SkNx(const __m128d& vec) : fVec(vec) {}
+
+    SkNx() {}
+    SkNx(double val) : fVec(_mm_set1_pd(val)) {}
+    static SkNx Load(const double vals[2]) { return _mm_loadu_pd(vals); }
+    SkNx(double a, double b) : fVec(_mm_setr_pd(a,b)) {}
+
+    void store(double vals[2]) const { _mm_storeu_pd(vals, fVec); }
+
+    SkNx operator + (const SkNx& o) const { return _mm_add_pd(fVec, o.fVec); }
+    SkNx operator - (const SkNx& o) const { return _mm_sub_pd(fVec, o.fVec); }
+    SkNx operator * (const SkNx& o) const { return _mm_mul_pd(fVec, o.fVec); }
+    SkNx operator / (const SkNx& o) const { return _mm_div_pd(fVec, o.fVec); }
+
+    SkNx operator == (const SkNx& o) const { return _mm_cmpeq_pd (fVec, o.fVec); }
+    SkNx operator != (const SkNx& o) const { return _mm_cmpneq_pd(fVec, o.fVec); }
+    SkNx operator  < (const SkNx& o) const { return _mm_cmplt_pd (fVec, o.fVec); }
+    SkNx operator  > (const SkNx& o) const { return _mm_cmpgt_pd (fVec, o.fVec); }
+    SkNx operator <= (const SkNx& o) const { return _mm_cmple_pd (fVec, o.fVec); }
+    SkNx operator >= (const SkNx& o) const { return _mm_cmpge_pd (fVec, o.fVec); }
+
+    static SkNx Min(const SkNx& l, const SkNx& r) { return _mm_min_pd(l.fVec, r.fVec); }
+    static SkNx Max(const SkNx& l, const SkNx& r) { return _mm_max_pd(l.fVec, r.fVec); }
+
+    SkNx sqrt() const { return _mm_sqrt_pd(fVec);  }
+
+    template <int k> double kth() const {
+        SkASSERT(0 <= k && k < 2);
+        union { __m128d v; double fs[2]; } pun = {fVec};
+        return pun.fs[k&1];
+    }
+
+    bool allTrue() const { return 0x3 == _mm_movemask_pd(fVec); }
+    bool anyTrue() const { return 0x0 != _mm_movemask_pd(fVec); }
+
+    SkNx thenElse(const SkNx& t, const SkNx& e) const {
+        return _mm_or_pd(_mm_and_pd   (fVec, t.fVec),
+                         _mm_andnot_pd(fVec, e.fVec));
+    }
+
+    __m128d fVec;
+};
+
+template <>
 class SkNx<4, int> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(int val) : fVec(_mm_set1_epi32(val)) {}
     static SkNx Load(const int vals[4]) { return _mm_loadu_si128((const __m128i*)vals); }
     SkNx(int a, int b, int c, int d) : fVec(_mm_setr_epi32(a,b,c,d)) {}
 
     void store(int vals[4]) const { _mm_storeu_si128((__m128i*)vals, fVec); }
@@ skipping 250 matching lines @@
                      _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))));
 }
 
 
 }  // namespace
 
 #endif//SkNx_sse_DEFINED