aacc + bbdd

src/opts/Sk4x_sse.h

 // It is important _not_ to put header guards here.
 // This file will be intentionally included three times.
 
 // Useful reading:
 //   https://software.intel.com/sites/landingpage/IntrinsicsGuide/
 
 #include "SkTypes.h"  // Keep this before any #ifdef for skbug.com/3362
 
 #if defined(SK4X_PREAMBLE)
     // Code in this file may assume SSE and SSE2.
@@ skipping 44 matching lines @@
 template <typename T> Sk4x<T>::Sk4x(const Sk4x& other) { *this = other; }
 template <typename T> Sk4x<T>& Sk4x<T>::operator=(const Sk4x<T>& other) {
     fVec = other.fVec;
     return *this;
 }
 
 // We pun in these _mm_shuffle_* methods a little to use the fastest / most available methods.
 // They're all bit-preserving operations so it shouldn't matter.
 
 template <typename T>
+Sk4x<T> Sk4x<T>::aacc() const { return _mm_shuffle_epi32(as_4i(fVec), _MM_SHUFFLE(2,2,0,0)); }
+template <typename T>
+Sk4x<T> Sk4x<T>::bbdd() const { return _mm_shuffle_epi32(as_4i(fVec), _MM_SHUFFLE(3,3,1,1)); }
+template <typename T>
 Sk4x<T> Sk4x<T>::badc() const { return _mm_shuffle_epi32(as_4i(fVec), _MM_SHUFFLE(2,3,0,1)); }
 
 // Now we'll write all Sk4f specific methods.  This M() macro will remove some noise.
 #define M(...) template <> inline __VA_ARGS__ Sk4f::
 
 M() Sk4x(float v) : fVec(_mm_set1_ps(v)) {}
 M() Sk4x(float a, float b, float c, float d) : fVec(_mm_set_ps(d,c,b,a)) {}
 
 M(Sk4f) Load       (const float fs[4]) { return _mm_loadu_ps(fs); }
 M(Sk4f) LoadAligned(const float fs[4]) { return _mm_load_ps (fs); }
@@ skipping 89 matching lines @@
     }
     M(Sk4i) Max(const Sk4i& a, const Sk4i& b) {
         Sk4i less = a.lessThan(b);
         return b.bitAnd(less).bitOr(a.andNot(less));
     }
 #endif
 
 #undef M
 
 #endif//Method definitions.