Implement SSE2-based implementations of the morphology filters (dilate &

src/opts/SkMorphology_opts_SSE2.cpp

Index: src/opts/SkMorphology_opts_SSE2.cpp
diff --git a/src/opts/SkMorphology_opts_SSE2.cpp b/src/opts/SkMorphology_opts_SSE2.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e6cfab586842efc08dbbde40a874f929b47fd33f
--- /dev/null
+++ b/src/opts/SkMorphology_opts_SSE2.cpp
@@ -0,0 +1,70 @@
+/*
+ * Copyright 2013 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 "SkColorPriv.h"
+
+#include <emmintrin.h>
+
+/* SSE2 version of dilateX, dilateY, erodeX, erodeY.
+ * portable versions are in src/effects/SkMorphologyImageFilter.cpp.
+ */
+
+template<bool dilate, bool inX>
+static void SkMorph_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                         int width, int height, int srcStride, int dstStride)
+{
+    const int srcStrideX = inX ? 1 : srcStride;
+    const int dstStrideX = inX ? 1 : dstStride;
+    const int srcStrideY = inX ? srcStride : 1;
+    const int dstStrideY = inX ? dstStride : 1;
+    radius = SkMin32(radius, width - 1);
+    const SkPMColor* upperSrc = src + radius * srcStrideX;
+    for (int x = 0; x < width; ++x) {
+        const SkPMColor* lp = src;
+        const SkPMColor* up = upperSrc;
+        SkPMColor* dptr = dst;
+        for (int y = 0; y < height; ++y) {
+            __m128i max = dilate ? _mm_setzero_si128() : _mm_set1_epi32(0xFFFFFFFF);
+            for (const SkPMColor* p = lp; p <= up; p += srcStrideX) {
+                __m128i src_pixel = _mm_cvtsi32_si128(*p);
+                max = dilate ? _mm_max_epu8(src_pixel, max) : _mm_min_epu8(src_pixel, max);
+            }
+            *dptr = _mm_cvtsi128_si32(max);
+            dptr += dstStrideY;
+            lp += srcStrideY;
+            up += srcStrideY;
+        }
+        if (x >= radius) src += srcStrideX;
+        if (x + radius < width - 1) upperSrc += srcStrideX;
+        dst += dstStrideX;
+    }
+}
+
+void SkDilateX_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                    int width, int height, int srcStride, int dstStride)
+{
+    SkMorph_SSE2<true, true>(src, dst, radius, width, height, srcStride, dstStride);
+}
+
+void SkErodeX_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                   int width, int height, int srcStride, int dstStride)
+{
+    SkMorph_SSE2<false, true>(src, dst, radius, width, height, srcStride, dstStride);
+}
+
+void SkDilateY_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                    int width, int height, int srcStride, int dstStride)
+{
+    SkMorph_SSE2<true, false>(src, dst, radius, width, height, srcStride, dstStride);
+}
+
+void SkErodeY_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                   int width, int height, int srcStride, int dstStride)
+{
+    SkMorph_SSE2<false, false>(src, dst, radius, width, height, srcStride, dstStride);
+}