Revert of Combined approach. (patchset #2 id:20001 of https://codereview.chromium.org/1356133002/ )

Reason for revert:
whee

Original issue's description:
> Combined approach.
>
> This combines some ideas from these two CLs:
>     - try stosd/w
>     - update memset16/32 inlining heuristics
>
>
> BUG=skia:4316
>
> Blinking in and out for perf.skia.org.
> TBR=reed@google.com
>
> Committed: https://skia.googlesource.com/skia/+/46243a7c02a1d5116e55a27ff59218f9c320df97

TBR=reed@google.com,mtklein@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:4316

Review URL: https://codereview.chromium.org/1353703006

1 files changed, 13 insertions, 29 deletions

diff --git a/include/core/SkUtils.h b/include/core/SkUtils.h
index 652cdae36e..4e24bd0883 100644
--- a/include/core/SkUtils.h
+++ b/include/core/SkUtils.h
@@ -9,27 +9,6 @@
 #define SkUtils_DEFINED
 
 #include "SkTypes.h"
-#if defined(SK_BUILD_FOR_WIN)
-    #include <intrin.h>
-#endif
-
-#if defined(SK_CPU_X86)
-    static inline void rep_stosw(uint16_t buffer[], uint16_t value, int count) {
-    #if defined(SK_BUILD_FOR_WIN)
-        __stosw(buffer, value, count);
-    #else
-        __asm__ __volatile__ ( "rep stosw" : "+D"(buffer), "+c"(count) : "a"(value) );
-    #endif
-    }
-
-    static inline void rep_stosd(uint32_t buffer[], uint32_t value, int count) {
-    #if defined(SK_BUILD_FOR_WIN)
-        __stosd((PDWORD)buffer, value, count);
-    #else
-        __asm__ __volatile__ ( "rep stosl" : "+D"(buffer), "+c"(count) : "a"(value) );
-    #endif
-    }
-#endif
 
 namespace SkOpts {
     extern void (*memset16)(uint16_t[], uint16_t, int);
@@ -38,8 +17,13 @@ namespace SkOpts {
 
 ///////////////////////////////////////////////////////////////////////////////
 
-// The stosw/d and inlining heuristics below were determined using
-// bench/MemsetBench.cpp and perf.skia.org.
+// The inlining heuristics below were determined using bench/MemsetBench.cpp
+// on a x86 desktop, a Nexus 7 with and without NEON, and a Nexus 9:
+//   - on x86, inlining was never faster,
+//   - on ARMv7, inlining was faster for N<=10.  Putting this check inside the NEON
+//     code was not helpful; it's got to be here outside.
+//   - NEON code generation for ARMv8 with GCC 4.9 is terrible,
+//     making the NEON code ~8x slower that just a serial loop.
 
 /** Similar to memset(), but it assigns a 16bit value into the buffer.
     @param buffer   The memory to have value copied into it
@@ -47,10 +31,10 @@ namespace SkOpts {
     @param count    The number of times value should be copied into the buffer.
 */
 static inline void sk_memset16(uint16_t buffer[], uint16_t value, int count) {
-#if defined(SK_CPU_X86)
-    if (count > 30) { rep_stosw(buffer, value, count); return; }
-#elif defined(SK_ARM_HAS_NEON)
+#if defined(SK_CPU_ARM64)
     while (count --> 0) { *buffer++ = value; } return;
+#elif defined(SK_CPU_ARM32)
+    if (count <= 10) { while (count --> 0) { *buffer++ = value; } return; }
 #endif
     SkOpts::memset16(buffer, value, count);
 }
@@ -61,10 +45,10 @@ static inline void sk_memset16(uint16_t buffer[], uint16_t value, int count) {
     @param count    The number of times value should be copied into the buffer.
 */
 static inline void sk_memset32(uint32_t buffer[], uint32_t value, int count) {
-#if defined(SK_CPU_X86)
-    if (count > 30) { rep_stosd(buffer, value, count); return; }
-#elif defined(SK_ARM_HAS_NEON)
+#if defined(SK_CPU_ARM64)
     while (count --> 0) { *buffer++ = value; } return;
+#elif defined(SK_CPU_ARM32)
+    if (count <= 10) { while (count --> 0) { *buffer++ = value; } return; }
 #endif
     SkOpts::memset32(buffer, value, count);
 }