Improve SkARGB32_A8_BlitMask_SSE2

src/opts/SkBlitRow_opts_SSE2.cpp

 /*
  * Copyright 2012 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.
  */
 
 #include <emmintrin.h>
 #include "SkBitmapProcState_opts_SSE2.h"
 #include "SkBlitRow_opts_SSE2.h"
@@ skipping 423 matching lines @@
             __m128i *d = reinterpret_cast<__m128i*>(dst);
             __m128i rb_mask = _mm_set1_epi32(0x00FF00FF);
             __m128i c_256 = _mm_set1_epi16(256);
             __m128i c_1 = _mm_set1_epi16(1);
             __m128i src_pixel = _mm_set1_epi32(color);
             while (count >= 4) {
                 // Load 4 pixels each of src and dest.
                 __m128i dst_pixel = _mm_load_si128(d);
 
                 //set the aphla value
-                __m128i src_scale_wide =  _mm_set_epi8(0, *(mask+3),\
-                                0, *(mask+3),0, \
-                                *(mask+2),0, *(mask+2),\
-                                0,*(mask+1), 0,*(mask+1),\
-                                0, *mask,0,*mask);
+                __m128i src_scale_wide = _mm_cvtsi32_si128(*reinterpret_cast<const uint32_t*>(mask));

[mtklein, 2014/10/03 19:49:50]

Please replace tabs with spaces before landing this.

+                src_scale_wide = _mm_unpacklo_epi8(src_scale_wide,
+                                                   _mm_setzero_si128());
+                src_scale_wide = _mm_unpacklo_epi16(src_scale_wide, src_scale_wide);
 
                 //call SkAlpha255To256()
                 src_scale_wide = _mm_add_epi16(src_scale_wide, c_1);
 
                 // Get red and blue pixels into lower byte of each word.
                 __m128i dst_rb = _mm_and_si128(rb_mask, dst_pixel);
                 __m128i src_rb = _mm_and_si128(rb_mask, src_pixel);
 
                 // Get alpha and green into lower byte of each word.
                 __m128i dst_ag = _mm_srli_epi16(dst_pixel, 8);
@@ skipping 893 matching lines @@
                 uint32_t dst_expanded = SkExpand_rgb_16(*dst);
                 dst_expanded = dst_expanded * (SkAlpha255To256(255 - a) >> 3);
                 // now src and dst expanded are in g:11 r:10 x:1 b:10
                 *dst = SkCompact_rgb_16((src_expanded + dst_expanded) >> 5);
             }
             dst += 1;
             DITHER_INC_X(x);
         } while (--count != 0);
     }
 }