try to speed-up maprect + round2i + contains

include/private/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 264 matching lines @@
     void store(void* ptr) const { _mm_storeu_si128((__m128i*)ptr, fVec); }
 
     SkNx saturatedAdd(const SkNx& o) const { return _mm_adds_epu8(fVec, o.fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_epi8(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_epi8(fVec, o.fVec); }
 
     static SkNx Min(const SkNx& a, const SkNx& b) { return _mm_min_epu8(a.fVec, b.fVec); }
     SkNx operator < (const SkNx& o) const {
         // There's no unsigned _mm_cmplt_epu8, so we flip the sign bits then use a signed compare.
         auto flip = _mm_set1_epi8(char(0x80));

[bungeman-skia, 2016/07/11 22:07:49]

Heh, maybe something like the following would keep msvc happy? Too bad
_mm_set1_epi8 takes char instead of 'signed char' or int8_t or something sane.

#include <climits>

#if CHAR_MAX == UCHAR_MAX
        auto flip = _mm_set1_epi8(char(0x80));
#else
        auto flip = _mm_set1_epi8(char(-0x80));
#endif

         return _mm_cmplt_epi8(_mm_xor_si128(flip, fVec), _mm_xor_si128(flip, o.fVec));
     }
 
     uint8_t operator[](int k) const {
         SkASSERT(0 <= k && k < 16);
         union { __m128i v; uint8_t us[16]; } pun = {fVec};
         return pun.us[k&15];
     }
 
     SkNx thenElse(const SkNx& t, const SkNx& e) const {
@@ skipping 69 matching lines @@
 
 template<> /*static*/ inline Sk4h SkNx_cast<uint16_t, uint8_t>(const Sk4b& src) {
     return _mm_unpacklo_epi8(src.fVec, _mm_setzero_si128());
 }
 
 template<> /*static*/ inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) {
     return _mm_packus_epi16(src.fVec, src.fVec);
 }
 
 #endif//SkNx_sse_DEFINED