(SSE2) acceleration for rectangular opaque erases.

15% speedup for rectangles < 31 px wide, 5% for larger.

http://codereview.appspot.com/5843050/



git-svn-id: http://skia.googlecode.com/svn/trunk@3423 2bbb7eff-a529-9590-31e7-b0007b416f81

3 files changed, 30 insertions, 4 deletions

diff --git a/src/core/SkBlitRow_D32.cpp b/src/core/SkBlitRow_D32.cpp
index 97aa665cb9..f1bf0ca1e1 100644
--- a/src/core/SkBlitRow_D32.cpp
+++ b/src/core/SkBlitRow_D32.cpp
@@ -12,6 +12,8 @@
 
 #define UNROLL
 
+SkBlitRow::ColorRectProc PlatformColorRectProcFactory();
+
 static void S32_Opaque_BlitRow32(SkPMColor* SK_RESTRICT dst,
                                  const SkPMColor* SK_RESTRICT src,
                                  int count, U8CPU alpha) {
@@ -178,3 +180,21 @@ void SkBlitRow::Color32(SkPMColor* SK_RESTRICT dst,
     }
 }
 
+void SkBlitRow::ColorRect32(SkPMColor* dst, int width, int height,
+                            size_t rowBytes, SkPMColor color) {
+    SkBlitRow::ColorProc proc = SkBlitRow::ColorProcFactory();
+    while (--height >= 0) {
+        (*proc)(dst, dst, width, color);
+        dst = (SkPMColor*) ((char*)dst + rowBytes);
+    }
+}
+
+SkBlitRow::ColorRectProc SkBlitRow::ColorRectProcFactory() {
+    SkBlitRow::ColorRectProc proc = PlatformColorRectProcFactory();
+    if (NULL == proc) {
+        proc = ColorRect32;
+    }
+    SkASSERT(proc);
+    return proc;
+}
+
diff --git a/src/core/SkBlitter_ARGB32.cpp b/src/core/SkBlitter_ARGB32.cpp
index 24ab330769..977c961583 100644
--- a/src/core/SkBlitter_ARGB32.cpp
+++ b/src/core/SkBlitter_ARGB32.cpp
@@ -53,6 +53,7 @@ SkARGB32_Blitter::SkARGB32_Blitter(const SkBitmap& device, const SkPaint& paint)
 
     fPMColor = SkPackARGB32(fSrcA, fSrcR, fSrcG, fSrcB);
     fColor32Proc = SkBlitRow::ColorProcFactory();
+    fColorRect32Proc = SkBlitRow::ColorRectProcFactory();
 }
 
 const SkBitmap* SkARGB32_Blitter::justAnOpaqueColor(uint32_t* value) {
@@ -213,10 +214,14 @@ void SkARGB32_Blitter::blitRect(int x, int y, int width, int height) {
     uint32_t    color = fPMColor;
     size_t      rowBytes = fDevice.rowBytes();
 
-    while (--height >= 0) {
-        fColor32Proc(device, device, width, color);
-        device = (uint32_t*)((char*)device + rowBytes);
-    }
+    //if (255 == SkGetPackedA32(color)) {
+        fColorRect32Proc(device, width, height, rowBytes, color);
+    //} else {
+        //while (--height >= 0) {
+            //fColor32Proc(device, device, width, color);
+            //device = (uint32_t*)((char*)device + rowBytes);
+        //}
+    //}
 }
 
 #if defined _WIN32 && _MSC_VER >= 1300
diff --git a/src/core/SkCoreBlitters.h b/src/core/SkCoreBlitters.h
index 4947198bcd..4a03a53169 100644
--- a/src/core/SkCoreBlitters.h
+++ b/src/core/SkCoreBlitters.h
@@ -94,6 +94,7 @@ protected:
     SkColor                fColor;
     SkPMColor              fPMColor;
     SkBlitRow::ColorProc   fColor32Proc;
+    SkBlitRow::ColorRectProc fColorRect32Proc;
 
 private:
     unsigned fSrcA, fSrcR, fSrcG, fSrcB;