Improve SkARGB32_A8_BlitMask_SSE2

With clang this:

-       movzbl  -3(%rbx), %edx
-       pxor    %xmm5, %xmm5
-       pinsrw  $0, %edx, %xmm5
-       pinsrw  $1, %edx, %xmm5
-       movzbl  -2(%rbx), %edx
-       pinsrw  $2, %edx, %xmm5
-       pinsrw  $3, %edx, %xmm5
-       movzbl  -1(%rbx), %edx
-       pinsrw  $4, %edx, %xmm5
-       pinsrw  $5, %edx, %xmm5
-       movzbl  (%rbx), %edx
-       pinsrw  $6, %edx, %xmm5
-       pinsrw  $7, %edx, %xmm5

becomes:

+       movd    (%rbx), %xmm4
+       punpcklbw       %xmm9, %xmm4
+       punpcklwd       %xmm4, %xmm4

And clang already does better codegen than msvc 2013 on this.

BUG=skia:

Review URL: https://codereview.chromium.org/609823003

1 files changed, 4 insertions, 5 deletions

diff --git a/src/opts/SkBlitRow_opts_SSE2.cpp b/src/opts/SkBlitRow_opts_SSE2.cpp
index 391b24c867..363cdab9f0 100644
--- a/src/opts/SkBlitRow_opts_SSE2.cpp
+++ b/src/opts/SkBlitRow_opts_SSE2.cpp
@@ -441,11 +441,10 @@ void SkARGB32_A8_BlitMask_SSE2(void* device, size_t dstRB, const void* maskPtr,
                 __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));
+                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);