Rewrite memset benches, then use results to add a small-N optimization.

include/core/SkUtils.h

Index: include/core/SkUtils.h
diff --git a/include/core/SkUtils.h b/include/core/SkUtils.h
index d522ae0dea139555f6e8790ac06d8287e5771fd7..c0e91eef9ff9f0a40b3d3c0d72632f8ecf4c2c3a 100644
--- a/include/core/SkUtils.h
+++ b/include/core/SkUtils.h
@@ -17,7 +17,18 @@
     @param value    The 16bit value to be copied into buffer
     @param count    The number of times value should be copied into the buffer.
 */
-void sk_memset16(uint16_t dst[], uint16_t value, int count);
+void sk_memset16_large(uint16_t dst[], uint16_t value, int count);

[reed1, 2015/04/09 18:55:48]

Seems like its not much harder to provide both N values...

// These values were determined running ...
#ifdef SK_BUILD_FOR_ARM && NOT_64BIT
    #define kMEMSET_SMALL_CUTOFF    10
#else
    #define kMEMSET_SMALL_CUTOFF    10000
#endif

[mtklein, 2015/04/09 19:11:10]

Good point!  Done.

+inline void sk_memset16(uint16_t dst[], uint16_t value, int count) {
+    // 10 determined empirically using bench/MemsetBench.cpp on a Nexus 7, Nexus 9, and desktop.
+    // (N7 was the limiting factor; desktops and N9 consider 10,000 small but 100,000 large).
+    if (count <= 10) {
+        for (int i = 0; i < count; i++) {
+            dst[i] = value;
+        }
+    } else {
+        sk_memset16_large(dst, value, count);
+    }
+}
 typedef void (*SkMemset16Proc)(uint16_t dst[], uint16_t value, int count);
 SkMemset16Proc SkMemset16GetPlatformProc();
 
@@ -26,7 +37,19 @@ SkMemset16Proc SkMemset16GetPlatformProc();
     @param value    The 32bit value to be copied into buffer
     @param count    The number of times value should be copied into the buffer.
 */
-void sk_memset32(uint32_t dst[], uint32_t value, int count);
+void sk_memset32_large(uint32_t dst[], uint32_t value, int count);
+inline void sk_memset32(uint32_t dst[], uint32_t value, int count) {
+    // 10 determined empirically using bench/MemsetBench.cpp on a Nexus 7, Nexus 9, and desktop.
+    // (N7 was the limiting factor; desktops and N9 consider 10,000 small but 100,000 large).
+    if (count <= 10) {
+        for (int i = 0; i < count; i++) {
+            dst[i] = value;
+        }
+    } else {
+        sk_memset32_large(dst, value, count);
+    }
+}
+
 typedef void (*SkMemset32Proc)(uint32_t dst[], uint32_t value, int count);
 SkMemset32Proc SkMemset32GetPlatformProc();