Rewriting MatrixConvolution image filter with SSE and AVX2

src/core/SkOpts.cpp

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkHalf.h"
 #include "SkOnce.h"
 #include "SkOpts.h"
 
 #define SK_OPTS_NS sk_default
 #include "SkBlitMask_opts.h"
 #include "SkBlitRow_opts.h"
 #include "SkBlurImageFilter_opts.h"
 #include "SkColorCubeFilter_opts.h"
 #include "SkMatrix_opts.h"
 #include "SkMorphologyImageFilter_opts.h"
 #include "SkSwizzler_opts.h"
 #include "SkTextureCompressor_opts.h"
 #include "SkXfermode_opts.h"
+#include "SkMatrixConvolutionImageFilter_opts.h"
 
 namespace SK_OPTS_NS {
     static void float_to_half(uint16_t dst[], const float src[], int n) {
         while (n-->0) {
             *dst++ = SkFloatToHalf(*src++);
         }
     }
     static void half_to_float(float dst[], const uint16_t src[], int n) {
         while (n-->0) {
             *dst++ = SkHalfToFloat(*src++);
@@ skipping 67 matching lines @@
     decltype(RGB_to_BGR1)           RGB_to_BGR1           = sk_default::RGB_to_BGR1;
     decltype(gray_to_RGB1)          gray_to_RGB1          = sk_default::gray_to_RGB1;
     decltype(grayA_to_RGBA)         grayA_to_RGBA         = sk_default::grayA_to_RGBA;
     decltype(grayA_to_rgbA)         grayA_to_rgbA         = sk_default::grayA_to_rgbA;
     decltype(inverted_CMYK_to_RGB1) inverted_CMYK_to_RGB1 = sk_default::inverted_CMYK_to_RGB1;
     decltype(inverted_CMYK_to_BGR1) inverted_CMYK_to_BGR1 = sk_default::inverted_CMYK_to_BGR1;
 
     decltype(half_to_float) half_to_float = sk_default::half_to_float;
     decltype(float_to_half) float_to_half = sk_default::float_to_half;
 
+    decltype(matrix_convolution_image_filter_filter_pixels) matrix_convolution_image_filter_filter_pixels = sk_default::SkMatrixConvolutionImageFilter_filterPixels_none;
+
     // Each Init_foo() is defined in src/opts/SkOpts_foo.cpp.
+    void Init_sse_sse2();
     void Init_ssse3();
     void Init_sse41();
     void Init_sse42() {}
     void Init_avx() {}
-    void Init_avx2() {}
+    void Init_avx2();
     void Init_neon();
 
     static void init() {
         // TODO: Chrome's not linking _sse* opts on iOS simulator builds.  Bug or feature?
     #if defined(SK_CPU_X86) && !defined(SK_BUILD_FOR_IOS)
+
+        // 'opts' target for x86 is already compiled with -msse2. Therefore, MMX, SSE, SSE2
+        // should be supported by every x86 CPU. So, just call Init_sse_sse2 function without
+        // any additional check.
+        Init_sse_sse2();
+
         uint32_t abcd[] = {0,0,0,0};
         cpuid(abcd);
+
         if (abcd[2] & (1<< 9)) { Init_ssse3(); }
         if (abcd[2] & (1<<19)) { Init_sse41(); }
         if (abcd[2] & (1<<20)) { Init_sse42(); }
 
         // AVX detection's kind of a pain.  This is cribbed from Chromium.
         if ( (  abcd[2] & (7<<26)) == (7<<26) &&    // Check bits 26-28 of ecx are all set,
              (xgetbv(0) & 6      ) == 6          ){ // and  check the OS supports XSAVE.
             Init_avx();
 
             // AVX2 additionally needs bit 5 set on ebx after calling cpuid(7).
@@ skipping 12 matching lines @@
 
     SK_DECLARE_STATIC_ONCE(gInitOnce);
     void Init() { SkOnce(&gInitOnce, init); }
 
 #if SK_ALLOW_STATIC_GLOBAL_INITIALIZERS
     static struct AutoInit {
         AutoInit() { Init(); }
     } gAutoInit;
 #endif
 }