Expand _01 half<->float limitation to _finite. Simplify.

src/opts/SkNx_neon.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_neon_DEFINED
 #define SkNx_neon_DEFINED
 
@@ skipping 370 matching lines @@
         union { int32x4_t v; int is[4]; } pun = {fVec};
         return pun.is[k&3];
     }
 
     SkNx operator + (const SkNx& o) const { return vaddq_s32(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return vsubq_s32(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return vmulq_s32(fVec, o.fVec); }
 
     SkNx operator & (const SkNx& o) const { return vandq_s32(fVec, o.fVec); }
     SkNx operator | (const SkNx& o) const { return vorrq_s32(fVec, o.fVec); }
+    SkNx operator ^ (const SkNx& o) const { return veorq_s32(fVec, o.fVec); }
 
     SkNx operator << (int bits) const { SHIFT32(vshlq_n_s32, fVec, bits); }
     SkNx operator >> (int bits) const { SHIFT32(vshrq_n_s32, fVec, bits); }
 
+    SkNx operator == (const SkNx& o) const {
+        return vreinterpretq_s32_u32(vceqq_s32(fVec, o.fVec));
+    }
+    SkNx operator <  (const SkNx& o) const {
+        return vreinterpretq_s32_u32(vcltq_s32(fVec, o.fVec));
+    }
+    SkNx operator >  (const SkNx& o) const {
+        return vreinterpretq_s32_u32(vcgtq_s32(fVec, o.fVec));
+    }
+
     static SkNx Min(const SkNx& a, const SkNx& b) { return vminq_s32(a.fVec, b.fVec); }
     // TODO as needed
 
+    SkNx thenElse(const SkNx& t, const SkNx& e) const {
+        return vbslq_s32(vreinterpretq_u32_s32(fVec), t.fVec, e.fVec);
+    }
+
     int32x4_t fVec;
 };
 
 #undef SHIFT32
 #undef SHIFT16
 #undef SHIFT8
 
 template<> inline Sk4i SkNx_cast<int, float>(const Sk4f& src) {
     return vcvtq_s32_f32(src.fVec);
 
@@ skipping 41 matching lines @@
 
 template<> inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) {
     return vmovn_u16(vcombine_u16(src.fVec, src.fVec));
 }
 
 template<> inline Sk4b SkNx_cast<uint8_t, int>(const Sk4i& src) {
     uint16x4_t _16 = vqmovun_s32(src.fVec);
     return vqmovn_u16(vcombine_u16(_16, _16));
 }
 
+template<> inline Sk4i SkNx_cast<int, uint16_t>(const Sk4h& src) {
+    return vreinterpretq_s32_u32(vmovl_u16(src.fVec));
+}
+
+template<> inline Sk4h SkNx_cast<uint16_t, int>(const Sk4i& src) {
+    return vmovn_u32(vreinterpretq_u32_s32(src.fVec));
+}
+
 static inline Sk4i Sk4f_round(const Sk4f& x) {
     return vcvtq_s32_f32((x + 0.5f).fVec);
 }
 
 #endif//SkNx_neon_DEFINED