Move immintrin/arm_neon includes to where they are used.

src/core/SkSharedMutex.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkSharedLock_DEFINED
 #define SkSharedLock_DEFINED
 
 #include "SkAtomics.h"
 #include "SkSemaphore.h"
 #include "SkTypes.h"
 
 #ifdef SK_DEBUG
     #include "SkMutex.h"
+
+    // On GCC 4.8, targeting ARMv7 with NEON, using libc++, we need to typedef float float32_t,
+    // (or include <arm_neon.h> which does that) before #including <memory> here.
+    // This makes no sense.  I'm not very interested in understanding why... this is an old,
+    // bizarre platform configuration that we should just let die.
+    #include <ciso646>  // Include something innocuous to define _LIBCPP_VERISON if it's libc++.
+    #if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ == 8 \
+     && defined(SK_CPU_ARM32) && defined(SK_ARM_HAS_NEON) \
+     && defined(_LIBCPP_VERSION)
+        typedef float float32_t;
+    #endif
+
     #include <memory>
 #endif  // SK_DEBUG
 
 // There are two shared lock implementations one debug the other is high performance. They implement
 // an interface similar to pthread's rwlocks.
 // This is a shared lock implementation similar to pthreads rwlocks. The high performance
 // implementation is cribbed from Preshing's article:
 // http://preshing.com/20150316/semaphores-are-surprisingly-versatile/
 //
 // This lock does not obey strict queue ordering. It will always alternate between readers and
@@ skipping 46 matching lines @@
 public:
     SkAutoSharedMutexShared(SkSharedMutex& lock) : fLock(lock) { lock.acquireShared(); }
     ~SkAutoSharedMutexShared() { fLock.releaseShared(); }
 private:
     SkSharedMutex& fLock;
 };
 
 #define SkAutoSharedMutexShared(...) SK_REQUIRE_LOCAL_VAR(SkAutoSharedMutexShared)
 
 #endif // SkSharedLock_DEFINED