Use a spinlock in SkOnce.

src/core/SkOnce.h

 /*
  * Copyright 2013 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkOnce_DEFINED
 #define SkOnce_DEFINED
 
@@ skipping 11 matching lines @@
 //     SkOnce(&once, set_up_my_singleton, &singleton);
 //     SkASSERT(NULL != singleton);
 //     return *singleton;
 // }
 //
 // OnceTest.cpp also should serve as a few other simple examples.
 
 #include "SkThread.h"
 #include "SkTypes.h"
 
-#ifdef SK_USE_POSIX_THREADS
-#  define SK_ONCE_INIT { false, { PTHREAD_MUTEX_INITIALIZER } }
-#else
-#  define SK_ONCE_INIT { false, SkBaseMutex() }
-#endif
-
+#define SK_ONCE_INIT { false, { 0, SkDEBUGCODE(0) } }
 #define SK_DECLARE_STATIC_ONCE(name) static SkOnceFlag name = SK_ONCE_INIT
 
 struct SkOnceFlag;  // If manually created, initialize with SkOnceFlag once = SK_ONCE_INIT
 
 template <typename Func, typename Arg>
 inline void SkOnce(SkOnceFlag* once, Func f, Arg arg);
 
 //  ----------------------  Implementation details below here. -----------------------------
 
 struct SkOnceFlag {
     bool done;
-    SkBaseMutex mutex;
+    SkSpinlock lock;
 };
 
 // TODO(bungeman, mtklein): move all these *barrier* functions to SkThread when refactoring lands.
 
 #ifdef SK_BUILD_FOR_WIN
-#include <intrin.h>
+#  include <intrin.h>
 inline static void compiler_barrier() {
     _ReadWriteBarrier();
 }
 #else
 inline static void compiler_barrier() {
     asm volatile("" : : : "memory");
 }
 #endif
 
 inline static void full_barrier_on_arm() {
 #ifdef SK_CPU_ARM
-#if SK_ARM_ARCH >= 7
+#  if SK_ARM_ARCH >= 7
     asm volatile("dmb" : : : "memory");
-#else
+#  else
     asm volatile("mcr p15, 0, %0, c7, c10, 5" : : "r" (0) : "memory");
-#endif
+#  endif
 #endif
 }
 
 // On every platform, we issue a compiler barrier to prevent it from reordering
 // code.  That's enough for platforms like x86 where release and acquire
 // barriers are no-ops.  On other platforms we may need to be more careful;
 // ARM, in particular, needs real code for both acquire and release.  We use a
 // full barrier, which acts as both, because that the finest precision ARM
 // provides.
 
 inline static void release_barrier() {
     compiler_barrier();
     full_barrier_on_arm();
 }
 
 inline static void acquire_barrier() {
     compiler_barrier();
     full_barrier_on_arm();
 }
 
 // We've pulled a pretty standard double-checked locking implementation apart
 // into its main fast path and a slow path that's called when we suspect the
 // one-time code hasn't run yet.
 
 // This is the guts of the code, called when we suspect the one-time code hasn't been run yet.
 // This should be rarely called, so we separate it from SkOnce and don't mark it as inline.
 // (We don't mind if this is an actual function call, but odds are it'll be inlined anyway.)
 template <typename Func, typename Arg>
 static void sk_once_slow(SkOnceFlag* once, Func f, Arg arg) {
-    const SkAutoMutexAcquire lock(once->mutex);
+    const SkAutoSpinlock lock(&once->lock);
     if (!once->done) {
         f(arg);
         // Also known as a store-store/load-store barrier, this makes sure that the writes
         // done before here---in particular, those done by calling f(arg)---are observable
         // before the writes after the line, *done = true.
         //
         // In version control terms this is like saying, "check in the work up
         // to and including f(arg), then check in *done=true as a subsequent change".
         //
         // We'll use this in the fast path to make sure f(arg)'s effects are
@@ skipping 34 matching lines @@
     //
     // The release barrier in sk_once_slow guaranteed that once->done = true
     // happens after f(arg), so by syncing to once->done = true here we're
     // forcing ourselves to also wait until the effects of f(arg) are readble.
     acquire_barrier();
 }
 
 #undef ANNOTATE_BENIGN_RACE
 
 #endif  // SkOnce_DEFINED