Replace SSE optimization of Color32A_D565

src/opts/SkBlitRow_opts_SSE4.cpp

 #include "SkBlitRow_opts_SSE4.h"
 
 // Some compilers can't compile SSSE3 or SSE4 intrinsics.  We give them stub methods.
 // The stubs should never be called, so we make them crash just to confirm that.
 #if SK_CPU_SSE_LEVEL < SK_CPU_SSE_LEVEL_SSE41
 void S32A_Opaque_BlitRow32_SSE4(SkPMColor* SK_RESTRICT, const SkPMColor* SK_RESTRICT, int, U8CPU) {
     sk_throw();
 }
 
 void Color32A_D565_SSE4(uint16_t dst[], SkPMColor src, int count, int x, int y) {
@@ skipping 48 matching lines @@
 
     // Wrap up the last <= 15 pixels.
     for (int i = count16*16; i < count; i++) {
         // This check is not really necessarily, but it prevents pointless autovectorization.
         if (src[i] & 0xFF000000) {
             dst[i] = SkPMSrcOver(src[i], dst[i]);
         }
     }
 }
 
-static inline uint16_t Color32A_D565_1x(uint16_t dst, unsigned scale, uint32_t src_expand) {
-    uint32_t dst_expand = SkExpand_rgb_16(dst) * scale;
-    return SkCompact_rgb_16((src_expand + dst_expand) >> 5);
-}
-
 void Color32A_D565_SSE4(uint16_t dst[], SkPMColor src, int count, int x, int y) {
     SkASSERT(count > 0);
 
     uint32_t src_expand = (SkGetPackedG32(src) << 24) |
                           (SkGetPackedR32(src) << 13) |
                           (SkGetPackedB32(src) << 2);
     unsigned scale = SkAlpha255To256(0xFF - SkGetPackedA32(src)) >> 3;
 
     // Check if we have enough pixels to run SIMD
     if (count >= (int)(8 + (((16 - (size_t)dst) & 0x0F) >> 1))) {
         __m128i* dst_wide;
         const __m128i src_expand_wide = _mm_set1_epi32(src_expand);
         const __m128i scale_wide = _mm_set1_epi32(scale);
         const __m128i mask_green = _mm_set1_epi32(SK_R16_MASK_IN_PLACE |
                                                   SK_B16_MASK_IN_PLACE |
                                                  (SK_G16_MASK_IN_PLACE << 16));
 
         // Align dst to an even 16 byte address (0-7 pixels)
         while (((((size_t)dst) & 0x0F) != 0) && (count > 0)) {
-            *dst = Color32A_D565_1x(*dst, scale, src_expand);
+            *dst = SkBlend32A_D565(*dst, scale, src_expand);
             dst += 1;
             count--;
         }
 
         dst_wide = reinterpret_cast<__m128i*>(dst);
         do {
             // Load 8 RGB565 pixels
             __m128i pixels = _mm_load_si128(dst_wide);
 
             // Duplicate and mask
@@ skipping 21 matching lines @@
             _mm_store_si128(dst_wide, pixels);
             count -= 8;
             dst_wide++;
         } while (count >= 8);
 
         dst = reinterpret_cast<uint16_t*>(dst_wide);
     }
 
     // Small loop to handle remaining pixels.
     while (count > 0) {
-        *dst = Color32A_D565_1x(*dst, scale, src_expand);
+        *dst = SkBlend32A_D565(*dst, scale, src_expand);
         dst += 1;
         count--;
     }
 }
 
 #endif