trim some fat from SSE2 fixed point alpha code

src/opts/Sk4px_SSE2.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.
  */
 
 namespace { // See Sk4px.h
 
 inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b(_mm_set1_epi32(px)); }
@@ skipping 45 matching lines @@
     }
 
     inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
         uint32_t as = *(const uint32_t*)a;
         __m128i splat = _mm_set_epi8(3,3,3,3, 2,2,2,2, 1,1,1,1, 0,0,0,0);
         return Sk16b(_mm_shuffle_epi8(_mm_cvtsi32_si128(as), splat));
     }
 #else
     inline Sk4px Sk4px::alphas() const {
         static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian.");
-        __m128i as = _mm_srli_epi32(this->fVec, 24);   // ___3 ___2 ___1 ___0
-        as = _mm_or_si128(as, _mm_slli_si128(as, 1));  // __33 __22 __11 __00
-        as = _mm_or_si128(as, _mm_slli_si128(as, 2));  // 3333 2222 1111 0000
+        // We exploit that A >= rgb for any premul pixel.

[msarett, 2015/11/17 19:59:57]

Clever :)

+        __m128i as = fVec;                             // 3xxx 2xxx 1xxx 0xxx
+        as = _mm_max_epu8(as, _mm_srli_epi32(as,  8)); // 33xx 22xx 11xx 00xx
+        as = _mm_max_epu8(as, _mm_srli_epi32(as, 16)); // 3333 2222 1111 0000
         return Sk16b(as);
     }
 
     inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
         __m128i as = _mm_cvtsi32_si128(*(const uint32_t*)a);  // ____ ____ ____ 3210
-        as = _mm_unpacklo_epi8 (as, _mm_setzero_si128());     // ____ ____ _3_2 _1_0
-        as = _mm_unpacklo_epi16(as, _mm_setzero_si128());     // ___3 ___2 ___1 ___0
-        as = _mm_or_si128(as, _mm_slli_si128(as, 1));         // __33 __22 __11 __00
-        as = _mm_or_si128(as, _mm_slli_si128(as, 2));         // 3333 2222 1111 0000
+        as = _mm_unpacklo_epi8 (as, as);                      // ____ ____ 3322 1100
+        as = _mm_unpacklo_epi16(as, as);                      // 3333 2222 1111 0000
         return Sk16b(as);
     }
 #endif
 
 inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
     uint32_t as = *(const uint16_t*)a;   // Aa -> Aa00
     return Load4Alphas((const SkAlpha*)&as);
 }
 
 inline Sk4px Sk4px::zeroColors() const {
@@ skipping 74 matching lines @@
     uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec));
     dst[0] = dst2;
     dst[1] = dst2 >> 16;
 }
 inline void Sk4px::store1(SkPMColor16 dst[1]) const {
     uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec));
     dst[0] = dst2;
 }
 
 }  // namespace