speed up rgn building by inlining memcmp for 32bit values

on mac/clang, using circularclips gm/bench

- before: 400ms
- after:  250ms

BUG=skia:
R=tomhudson@chromium.org, mtklein@google.com

Author: reed@google.com

Review URL: https://codereview.chromium.org/348143002

diff --git a/src/core/SkRegion_path.cpp b/src/core/SkRegion_path.cpp
index 03830e6ce542a1864f5a6ad144ad8b027c8090b7..108511c6c5d905eeb0df1186267a269b3e29bc8a 100644 (file)
--- a/src/core/SkRegion_path.cpp
+++ b/src/core/SkRegion_path.cpp
@@ -1,4 +1,3 @@
-
 /*
  * Copyright 2006 The Android Open Source Project
  *
@@ -6,13 +5,23 @@
  * found in the LICENSE file.
  */
 
-
 #include "SkRegionPriv.h"
 #include "SkBlitter.h"
 #include "SkScan.h"
 #include "SkTDArray.h"
 #include "SkPath.h"
 
+// The rgnbuilder caller *seems* to pass short counts, possible often seens early failure, so
+// we may not want to promote this to a "std" routine just yet.
+static bool sk_memeq32(const int32_t* SK_RESTRICT a, const int32_t* SK_RESTRICT b, int count) {
+    for (int i = 0; i < count; ++i) {
+        if (a[i] != b[i]) {
+            return false;
+        }
+    }
+    return true;
+}
+
 class SkRgnBuilder : public SkBlitter {
 public:
     SkRgnBuilder();
@@ -87,9 +96,7 @@ private:
         if (fPrevScanline != NULL &&
             fPrevScanline->fLastY + 1 == fCurrScanline->fLastY &&
             fPrevScanline->fXCount == fCurrScanline->fXCount &&
-            !memcmp(fPrevScanline->firstX(),
-                    fCurrScanline->firstX(),
-                    fCurrScanline->fXCount * sizeof(SkRegion::RunType)))
+            sk_memeq32(fPrevScanline->firstX(), fCurrScanline->firstX(), fCurrScanline->fXCount))
         {
             // update the height of fPrevScanline
             fPrevScanline->fLastY = fCurrScanline->fLastY;