Spin off just SkLazyFnPtr from 305513002.

The memory barrier in SkOnce is a perf regression for sk_mem{set,cpy} in
SkUtils on ARM.  We can do a lot better for function pointers.

BUG=skia:

Committed: http://code.google.com/p/skia/source/detail?r=14929

R=bungeman@google.com, mtklein@google.com

Author: mtklein@chromium.org

Review URL: https://codereview.chromium.org/305753002

git-svn-id: http://skia.googlecode.com/svn/trunk@14938 2bbb7eff-a529-9590-31e7-b0007b416f81

src/core/SkLazyFnPtr.h

@@ -0,0 +1,58 @@
+#ifndef SkLazyFnPtr_DEFINED
+#define SkLazyFnPtr_DEFINED
+
+/** Declare a lazily-chosen static function pointer of type F.
+ *
+ *  Example usage:
+ *
+ *  typedef int (*FooImpl)(int, int);
+ *
+ *  static FooImpl choose_foo() { return ... };
+ *
+ *  int Foo(int a, int b) {
+ *     SK_DECLARE_STATIC_LAZY_FN_PTR(FooImpl, choice);
+ *     return choice.get(choose_foo)(a, b);
+ *  }
+ *
+ *  You can think of SK_DECLARE_STATIC_LAZY_FN_PTR as a cheaper specialization of SkOnce.
+ *  There is no mutex, and in the fast path, no memory barriers are issued.
+ *
+ *  This must be used in a global or function scope, not as a class member.
+ */
+#define SK_DECLARE_STATIC_LAZY_FN_PTR(F, name) static Private::SkLazyFnPtr<F> name = { NULL }
+
+
+// Everything below here is private implementation details.  Don't touch, don't even look.
+
+#include "SkDynamicAnnotations.h"
+#include "SkThreadPriv.h"
+
+namespace Private {
+
+// This has no constructor and is link-time initialized, so its members must be public.
+template <typename F>
+struct SkLazyFnPtr {
+    F get(F (*choose)()) {
+        // First, try reading to see if it's already set.
+        F fn = (F)SK_ANNOTATE_UNPROTECTED_READ(fPtr);
+        if (fn != NULL) {
+            return fn;
+        }
+
+        // We think it's not already set.
+        fn = choose();
+
+        // No particular memory barriers needed; we're not guarding anything but the pointer itself.
+        F prev = (F)sk_atomic_cas(&fPtr, NULL, (void*)fn);
+
+        // If prev != NULL, someone snuck in and set fPtr concurrently.
+        // If prev == NULL, we did write fn to fPtr.
+        return prev != NULL ? prev : fn;
+    }
+
+    void* fPtr;
+};
+
+}  // namespace Private
+
+#endif//SkLazyFnPtr_DEFINED

src/core/SkThreadPriv.h

@@ -0,0 +1,23 @@
+/*
+ * Copyright 2014 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef SkThreadPriv_DEFINED
+#define SkThreadPriv_DEFINED
+
+#include "SkTypes.h"
+
+// SK_ATOMICS_PLATFORM_H must provide inline implementations for the following declarations.
+
+/** Atomic compare and set, for pointers.
+ *  If *addr == before, set *addr to after.  Always returns previous value of *addr.
+ *  This must act as a compiler barrier.
+ */
+static void* sk_atomic_cas(void** addr, void* before, void* after);
+
+#include SK_ATOMICS_PLATFORM_H
+
+#endif//SkThreadPriv_DEFINED

src/core/SkUtils.cpp

@@ -8,7 +8,7 @@
 
 
 #include "SkUtils.h"
-#include "SkOnce.h"
+#include "SkLazyFnPtr.h"
 
 #if 0
 #define assign_16_longs(dst, value)             \
@@ -113,52 +113,34 @@ static void sk_memcpy32_portable(uint32_t dst[], const uint32_t src[], int count
     memcpy(dst, src, count * sizeof(uint32_t));
 }
 
-static void choose_memset16(SkMemset16Proc* proc) {
-    *proc = SkMemset16GetPlatformProc();
-    if (NULL == *proc) {
-        *proc = &sk_memset16_portable;
-    }
+static SkMemset16Proc choose_memset16() {
+    SkMemset16Proc proc = SkMemset16GetPlatformProc();
+    return proc ? proc : sk_memset16_portable;
 }
 
 void sk_memset16(uint16_t dst[], uint16_t value, int count) {
-    SK_DECLARE_STATIC_ONCE(once);
-    static SkMemset16Proc proc = NULL;
-    SkOnce(&once, choose_memset16, &proc);
-    SkASSERT(proc != NULL);
-
-    return proc(dst, value, count);
+    SK_DECLARE_STATIC_LAZY_FN_PTR(SkMemset16Proc, choice);
+    return choice.get(choose_memset16)(dst, value, count);
 }
 
-static void choose_memset32(SkMemset32Proc* proc) {
-    *proc = SkMemset32GetPlatformProc();
-    if (NULL == *proc) {
-        *proc = &sk_memset32_portable;
-    }
+static SkMemset32Proc choose_memset32() {
+    SkMemset32Proc proc = SkMemset32GetPlatformProc();
+    return proc ? proc : sk_memset32_portable;
 }
 
 void sk_memset32(uint32_t dst[], uint32_t value, int count) {
-    SK_DECLARE_STATIC_ONCE(once);
-    static SkMemset32Proc proc = NULL;
-    SkOnce(&once, choose_memset32, &proc);
-    SkASSERT(proc != NULL);
-
-    return proc(dst, value, count);
+    SK_DECLARE_STATIC_LAZY_FN_PTR(SkMemset32Proc, choice);
+    return choice.get(choose_memset32)(dst, value, count);
 }
 
-static void choose_memcpy32(SkMemcpy32Proc* proc) {
-    *proc = SkMemcpy32GetPlatformProc();
-    if (NULL == *proc) {
-        *proc = &sk_memcpy32_portable;
-    }
+static SkMemcpy32Proc choose_memcpy32() {
+    SkMemcpy32Proc proc = SkMemcpy32GetPlatformProc();
+    return proc ? proc : sk_memcpy32_portable;
 }
 
 void sk_memcpy32(uint32_t dst[], const uint32_t src[], int count) {
-    SK_DECLARE_STATIC_ONCE(once);
-    static SkMemcpy32Proc proc = NULL;
-    SkOnce(&once, choose_memcpy32, &proc);
-    SkASSERT(proc != NULL);
-
-    return proc(dst, src, count);
+    SK_DECLARE_STATIC_LAZY_FN_PTR(SkMemcpy32Proc, choice);
+    return choice.get(choose_memcpy32)(dst, src, count);
 }
 
 ///////////////////////////////////////////////////////////////////////////////

src/ports/SkAtomics_sync.h

@@ -32,6 +32,12 @@ static inline __attribute__((always_inline)) bool sk_atomic_cas(int32_t* addr,
     return __sync_bool_compare_and_swap(addr, before, after);
 }
 
+static inline __attribute__((always_inline)) void* sk_atomic_cas(void** addr,
+                                                                 void* before,
+                                                                 void* after) {
+    return __sync_val_compare_and_swap(addr, before, after);
+}
+
 static inline __attribute__((always_inline)) void sk_membar_acquire__after_atomic_conditional_inc() { }
 
 #endif

src/ports/SkAtomics_win.h

@@ -40,6 +40,10 @@ static inline bool sk_atomic_cas(int32_t* addr, int32_t before, int32_t after) {
     return _InterlockedCompareExchange(reinterpret_cast<long*>(addr), after, before) == before;
 }
 
+static inline void* sk_atomic_cas(void** addr, void* before, void* after) {
+    return InterlockedCompareExchangePointer(addr, after, before);
+}
+
 static inline void sk_membar_acquire__after_atomic_conditional_inc() { }
 
 #endif