Unroll loops in SkBlurMask for speedup on Windows (benchmarks should see

15% on interpolated blurs, 5-10% on simple blurs).



git-svn-id: http://skia.googlecode.com/svn/trunk@2755 2bbb7eff-a529-9590-31e7-b0007b416f81

3 files changed, 16 insertions, 1 deletions

diff --git a/include/core/SkEndian.h b/include/core/SkEndian.h
index d7dd562cf1..3eb67dabfe 100644
--- a/include/core/SkEndian.h
+++ b/include/core/SkEndian.h
@@ -80,6 +80,19 @@ static inline void SkEndianSwap32s(uint32_t array[], int count) {
     #define SkEndian_SwapLE32(n)    SkEndianSwap32(n)
 #endif
 
+// When a bytestream is embedded in a 32-bit word, how far we need to
+// shift the word to extract each byte from the low 8 bits by anding with 0xff.
+#ifdef SK_CPU_LENDIAN
+    #define SkEndian_Byte0Shift 0
+    #define SkEndian_Byte1Shift 8
+    #define SkEndian_Byte2Shift 16
+    #define SkEndian_Byte3Shift 24
+#else   // SK_CPU_BENDIAN
+    #define SkEndian_Byte0Shift 24
+    #define SkEndian_Byte1Shift 16
+    #define SkEndian_Byte2Shift 8
+    #define SkEndian_Byte3Shift 0
+#endif
 
 #endif
 
diff --git a/include/core/SkTypes.h b/include/core/SkTypes.h
index 1ee25b129a..51d8d187c7 100644
--- a/include/core/SkTypes.h
+++ b/include/core/SkTypes.h
@@ -216,6 +216,8 @@ static inline bool SkIsU16(long x) {
 */
 #define SkAlign4(x)     (((x) + 3) >> 2 << 2)
 
+#define SkIsAlign4(x) (((x) & 3) == 0)
+
 typedef uint32_t SkFourByteTag;
 #define SkSetFourByteTag(a, b, c, d)    (((a) << 24) | ((b) << 16) | ((c) << 8) | (d))
 
diff --git a/include/gpu/GrTypes.h b/include/gpu/GrTypes.h
index eb92f09eb7..e961fac805 100644
--- a/include/gpu/GrTypes.h
+++ b/include/gpu/GrTypes.h
@@ -55,7 +55,7 @@
  *  n is already a multiple of 4
  */
 #define GrALIGN4(n)     SkAlign4(n)
-#define GrIsALIGN4(n)   (((n) & 3) == 0)
+#define GrIsALIGN4(n)   SkIsAlign4(n)
 
 template <typename T> const T& GrMin(const T& a, const T& b) {
 	return (a < b) ? a : b;