add Min to SkNi, specialized for u8 and u16 on SSE and NEON

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 254 matching lines @@
 
     void store(uint16_t vals[8]) const { _mm_storeu_si128((__m128i*)vals, fVec); }
 
     SkNi operator + (const SkNi& o) const { return _mm_add_epi16(fVec, o.fVec); }
     SkNi operator - (const SkNi& o) const { return _mm_sub_epi16(fVec, o.fVec); }
     SkNi operator * (const SkNi& o) const { return _mm_mullo_epi16(fVec, o.fVec); }
 
     SkNi operator << (int bits) const { return _mm_slli_epi16(fVec, bits); }
     SkNi operator >> (int bits) const { return _mm_srli_epi16(fVec, bits); }
 
+    static SkNi Min(const SkNi& a, const SkNi& b) {
+        // No unsigned _mm_min_epu16, so we'll shift into a space where we can use the
+        // signed version, _mm_min_epi16, then shift back.
+        const uint16_t top = 0x8000; // Keep this separate from _mm_set1_epi16 or MSVC will whine.
+        const __m128i top_8x = _mm_set1_epi16(top);
+        return _mm_add_epi8(top_8x, _mm_min_epi16(_mm_sub_epi8(a.fVec, top_8x),
+                                                  _mm_sub_epi8(b.fVec, top_8x)));
+    }
+
     template <int k> uint16_t kth() const {
         SkASSERT(0 <= k && k < 8);
         return _mm_extract_epi16(fVec, k);
     }
 
     __m128i fVec;
 };
 
 template <>
 class SkNi<16, uint8_t> {
@@ skipping 14 matching lines @@
     SkNi saturatedAdd(const SkNi& o) const { return _mm_adds_epu8(fVec, o.fVec); }
 
     SkNi operator + (const SkNi& o) const { return _mm_add_epi8(fVec, o.fVec); }
     SkNi operator - (const SkNi& o) const { return _mm_sub_epi8(fVec, o.fVec); }
 
     // SSE cannot multiply or shift vectors of uint8_t.
     SkNi operator * (const SkNi& o) const { SkASSERT(false); return fVec; }
     SkNi operator << (int bits) const { SkASSERT(false); return fVec; }
     SkNi operator >> (int bits) const { SkASSERT(false); return fVec; }
 
+    static SkNi Min(const SkNi& a, const SkNi& b) { return _mm_min_epu8(a.fVec, b.fVec); }
+
     template <int k> uint8_t kth() const {
         SkASSERT(0 <= k && k < 16);
         // SSE4.1 would just `return _mm_extract_epi8(fVec, k)`.  We have to read 16-bits instead.
         int pair = _mm_extract_epi16(fVec, k/2);
         return k % 2 == 0 ? pair : (pair >> 8);
     }
 
     __m128i fVec;
 };
 
 #endif//SkNx_sse_DEFINED