Optimize SkAlphaMulQ_SSE2

src/opts/SkColor_opts_SSE2.h

 /*
  * Copyright 2014 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkColor_opts_SSE2_DEFINED
 #define SkColor_opts_SSE2_DEFINED
 
 #include <emmintrin.h>
 
+#define ASSERT_EQ(a,b) SkASSERT(0xffff == _mm_movemask_epi8(_mm_cmpeq_epi8((a), (b))))
+
 // Because no _mm_mul_epi32() in SSE2, we emulate it here.
 // Multiplies 4 32-bit integers from a by 4 32-bit intergers from b.
 // The 4 multiplication results should be represented within 32-bit
 // integers, otherwise they would be overflow.
 static inline  __m128i Multiply32_SSE2(const __m128i& a, const __m128i& b) {
     // Calculate results of a0 * b0 and a2 * b2.
     __m128i r1 = _mm_mul_epu32(a, b);
     // Calculate results of a1 * b1 and a3 * b3.
     __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a, 4), _mm_srli_si128(b, 4));
     // Shuffle results to [63..0] and interleave the results.
@@ skipping 22 matching lines @@
     __m128i mask = _mm_set1_epi32(0xFF00FF);
     __m128i s = _mm_or_si128(_mm_slli_epi32(scale, 16), scale);
 
     // uint32_t rb = ((c & mask) * scale) >> 8
     __m128i rb = _mm_and_si128(mask, c);
     rb = _mm_mullo_epi16(rb, s);
     rb = _mm_srli_epi16(rb, 8);
 
     // uint32_t ag = ((c >> 8) & mask) * scale
     __m128i ag = _mm_srli_epi16(c, 8);
-    ag = _mm_and_si128(ag, mask);
+    ASSERT_EQ(ag, _mm_and_si128(mask, ag));  // ag = _mm_srli_epi16(c, 8) did this for us.
     ag = _mm_mullo_epi16(ag, s);
 
     // (rb & mask) | (ag & ~mask)
-    rb = _mm_and_si128(mask, rb);
+    ASSERT_EQ(rb, _mm_and_si128(mask, rb));  // rb = _mm_srli_epi16(rb, 8) did this for us.
     ag = _mm_andnot_si128(mask, ag);
     return _mm_or_si128(rb, ag);
 }
 
 static inline __m128i SkGetPackedA32_SSE2(const __m128i& src) {
     __m128i a = _mm_slli_epi32(src, (24 - SK_A32_SHIFT));
     return _mm_srli_epi32(a, 24);
 }
 
 static inline __m128i SkGetPackedR32_SSE2(const __m128i& src) {
@@ skipping 106 matching lines @@
                                 SK_B32_SHIFT + (8 - SK_B16_BITS));
     b2 = _mm_and_si128(b2, _mm_set1_epi32(SK_B16_MASK));
     __m128i b = _mm_packs_epi32(b1, b2);
 
     // Store 8 16-bit colors in dst.
     __m128i d_pixel = SkPackRGB16_SSE2(r, g, b);
 
     return d_pixel;
 }
 
+#undef ASSERT_EQ
 #endif // SkColor_opts_SSE2_DEFINED