Xfermode: SSE2 implementation of overlay_modeproc

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>
 
+static inline  __m128i Multiply32_SSE2(const __m128i& a, const __m128i& b) {
+    __m128i r1 = _mm_mul_epu32(a, b);
+    __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a, 4), _mm_srli_si128(b, 4));

[mtklein, 2014/04/22 15:18:21]

Can you narrate this method a bit more, at least with a comment at the top about
what math it does exactly on a and b?  I only ask because Multiply32_SSE2 is a
pretty generic name, and I can't find any corresponding non-SIMD code to compare
it with.

[qiankun, 2014/04/23 07:40:05]

Done.

+    __m128i r = _mm_unpacklo_epi32(_mm_shuffle_epi32(r1, _MM_SHUFFLE(0,0,2,0)),
+                                   _mm_shuffle_epi32(r2, _MM_SHUFFLE(0,0,2,0)));
+    return r;
+}
+
 static inline __m128i SkAlpha255To256_SSE2(const __m128i& alpha) {
     return _mm_add_epi32(alpha, _mm_set1_epi32(1));
 }
 
 // See #define SkAlphaMulAlpha(a, b)  SkMulDiv255Round(a, b) in SkXfermode.cpp.
 static inline __m128i SkAlphaMulAlpha_SSE2(const __m128i& a,
                                            const __m128i& b) {
     __m128i prod = _mm_mullo_epi16(a, b);
     prod = _mm_add_epi32(prod, _mm_set1_epi32(128));
     prod = _mm_add_epi32(prod, _mm_srli_epi32(prod, 8));
@@ skipping 139 matching lines @@
     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;
 }
 
 #endif // SkColor_opts_SSE2_DEFINED