aacc + bbdd

src/opts/Sk4x_neon.h

 // It is important _not_ to put header guards here.
 // This file will be intentionally included three times.
 
 #include "SkTypes.h"  // Keep this before any #ifdef for skbug.com/3362
 
 #if defined(SK4X_PREAMBLE)
     #include <arm_neon.h>
 
     // Template metaprogramming to map scalar types to vector types.
     template <typename T> struct SkScalarToSIMD;
@@ skipping 81 matching lines @@
 M(Sk4i) equal           (const Sk4f& o) const { return vreinterpretq_s32_u32(vceqq_f32(fVec, o.fVec)); }
 M(Sk4i) notEqual        (const Sk4f& o) const { return vreinterpretq_s32_u32(vmvnq_u32(vceqq_f32(fVec, o.fVec))); }
 M(Sk4i) lessThan        (const Sk4f& o) const { return vreinterpretq_s32_u32(vcltq_f32(fVec, o.fVec)); }
 M(Sk4i) greaterThan     (const Sk4f& o) const { return vreinterpretq_s32_u32(vcgtq_f32(fVec, o.fVec)); }
 M(Sk4i) lessThanEqual   (const Sk4f& o) const { return vreinterpretq_s32_u32(vcleq_f32(fVec, o.fVec)); }
 M(Sk4i) greaterThanEqual(const Sk4f& o) const { return vreinterpretq_s32_u32(vcgeq_f32(fVec, o.fVec)); }
 
 M(Sk4f) Min(const Sk4f& a, const Sk4f& b) { return vminq_f32(a.fVec, b.fVec); }
 M(Sk4f) Max(const Sk4f& a, const Sk4f& b) { return vmaxq_f32(a.fVec, b.fVec); }
 
+M(Sk4f) aacc() const { return vtrnq_f32(fVec, fVec).val[0]; }
+M(Sk4f) bbdd() const { return vtrnq_f32(fVec, fVec).val[1]; }
 M(Sk4f) badc() const { return vrev64q_f32(fVec); }
 
 // Sk4i Methods
 #undef M
 #define M(...) template <> inline __VA_ARGS__ Sk4i::
 
 M() Sk4x(int32_t v) : fVec(vdupq_n_s32(v)) {}
 M() Sk4x(int32_t a, int32_t b, int32_t c, int32_t d) { fVec = (int32x4_t) { a, b, c, d }; }
 
 // As far as I can tell, it's not possible to provide an alignment hint to
@@ skipping 36 matching lines @@
 M(Sk4i) lessThanEqual   (const Sk4i& o) const { return vreinterpretq_s32_u32(vcleq_s32(fVec, o.fVec)); }
 M(Sk4i) greaterThanEqual(const Sk4i& o) const { return vreinterpretq_s32_u32(vcgeq_s32(fVec, o.fVec)); }
 
 M(Sk4i) add     (const Sk4i& o) const { return vaddq_s32(fVec, o.fVec); }
 M(Sk4i) subtract(const Sk4i& o) const { return vsubq_s32(fVec, o.fVec); }
 M(Sk4i) multiply(const Sk4i& o) const { return vmulq_s32(fVec, o.fVec); }
 // NEON does not have integer reciprocal, sqrt, or division.
 M(Sk4i) Min(const Sk4i& a, const Sk4i& b) { return vminq_s32(a.fVec, b.fVec); }
 M(Sk4i) Max(const Sk4i& a, const Sk4i& b) { return vmaxq_s32(a.fVec, b.fVec); }
 
+M(Sk4i) aacc() const { return vtrnq_s32(fVec, fVec).val[0]; }
+M(Sk4i) bbdd() const { return vtrnq_s32(fVec, fVec).val[1]; }
 M(Sk4i) badc() const { return vrev64q_s32(fVec); }
 
 #undef M
 
 #endif