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..8d94edd5b95a94e18b285f698b03aa392876a0a7
--- /dev/null
+++ b/src/opts/SkMorphology_opts_SSE2.cpp
@@ -0,0 +1,117 @@
+/*
+ * 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.
+ */
+
+void SkDilateX_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,

[mtklein, 2013/10/30 15:44:47]

Do you see any value in sharing the structure of an X pass or a Y pass between
erode and dilate by parameterizing on the combiner (__mm_max_epu8,
__mm_min_epu8) and initial value (0 or 0xFFFFFFFF)?

I found myself doing a lot of paranoid double checking against the serial code
and across these dilate and erode here.

[Stephen White, 2013/10/30 19:47:17]

OK, I've templated across type and direction. No effect on performance (which is
good). Let me know what you think.

+                    int width, int height, int srcStride, int dstStride)
+{
+    radius = SkMin32(radius, width - 1);
+    const SkPMColor* upperSrc = src + radius;
+    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 = _mm_setzero_si128();
+            for (const SkPMColor* p = lp; p <= up; ++p) {
+                __m128i src_pixel = _mm_cvtsi32_si128(*p);
+                max = _mm_max_epu8(src_pixel, max);
+            }
+            *dptr = _mm_cvtsi128_si32(max);
+            dptr += dstStride;
+            lp += srcStride;
+            up += srcStride;
+        }
+        if (x >= radius) ++src;
+        if (x + radius < width - 1) ++upperSrc;
+        ++dst;
+    }
+}
+
+void SkDilateY_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                    int width, int height, int srcStride, int dstStride)
+{
+    radius = SkMin32(radius, height - 1);
+    const SkPMColor* upperSrc = src + radius * srcStride;
+    for (int y = 0; y < height; ++y) {
+        const SkPMColor* lp = src;
+        const SkPMColor* up = upperSrc;
+        SkPMColor* dptr = dst;
+        for (int x = 0; x < width; ++x) {
+            __m128i max = _mm_setzero_si128();
+            for (const SkPMColor* p = lp; p <= up; p += srcStride) {
+                __m128i src_pixel = _mm_cvtsi32_si128(*p);
+                max = _mm_max_epu8(src_pixel, max);
+            }
+            *dptr++ = _mm_cvtsi128_si32(max);
+            lp++;
+            up++;
+        }
+        if (y >= radius) src += srcStride;
+        if (y + radius < height - 1) upperSrc += srcStride;
+        dst += dstStride;
+    }
+}
+
+void SkErodeX_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                   int width, int height, int srcStride, int dstStride)
+{
+    radius = SkMin32(radius, width - 1);
+    const SkPMColor* upperSrc = src + radius;
+    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 min = _mm_set1_epi32(0xFFFFFFFF);
+            for (const SkPMColor* p = lp; p <= up; ++p) {
+                __m128i src_pixel = _mm_cvtsi32_si128(*p);
+                min = _mm_min_epu8(src_pixel, min);
+            }
+            *dptr = _mm_cvtsi128_si32(min);
+            dptr += dstStride;
+            lp += srcStride;
+            up += srcStride;
+        }
+        if (x >= radius) ++src;
+        if (x + radius < width - 1) ++upperSrc;
+        ++dst;
+    }
+}
+
+void SkErodeY_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+                   int width, int height, int srcStride, int dstStride)
+{
+    radius = SkMin32(radius, height - 1);
+    const SkPMColor* upperSrc = src + radius * srcStride;
+    for (int y = 0; y < height; ++y) {
+        const SkPMColor* lp = src;
+        const SkPMColor* up = upperSrc;
+        SkPMColor* dptr = dst;
+        for (int x = 0; x < width; ++x) {
+            __m128i min = _mm_set1_epi32(0xFFFFFFFF);
+            for (const SkPMColor* p = lp; p <= up; p += srcStride) {
+                __m128i src_pixel = _mm_cvtsi32_si128(*p);
+                min = _mm_min_epu8(src_pixel, min);
+            }
+            *dptr++ = _mm_cvtsi128_si32(min);
+            lp++;
+            up++;
+        }
+        if (y >= radius) src += srcStride;
+        if (y + radius < height - 1) upperSrc += srcStride;
+        dst += dstStride;
+    }
+}