Reduce dependence on C++ runtime

If the external thread statics are used, then
we don't need to include some C++ runtime
concepts. This refactoring moves some global initialization under
conditional compilation.

src/mem/largealloc.h

@@ -58,8 +58,39 @@ namespace snmalloc
   class MemoryProviderStateMixin : public PAL
   {
     std::atomic_flag lock = ATOMIC_FLAG_INIT;
-    void* bump;
-    size_t remaining;
+    void* bump = 0;
+    size_t remaining = 0;
+
+  public:
+    /**
+     * Stack of large allocations that have been returned for reuse.
+     */
+    ModArray<NUM_LARGE_CLASSES, MPMCStack<Largeslab, RequiresInit>> large_stack;
+
+  static MemoryProviderStateMixin<PAL>* make() noexcept
+  {
+    // Temporary storage to start the allocator in.
+    MemoryProviderStateMixin<PAL> local;
+
+    // Allocate permanent storage for the allocator usung temporary allocator
+    MemoryProviderStateMixin<PAL>* allocated 
+      = local.alloc_chunk<MemoryProviderStateMixin<PAL>, 1>();
+
+    // Put temporary allocator we have used, into the permanent storage.
+    // memcpy is safe as this is entirely single threaded.
+    memcpy(allocated, &local, sizeof(MemoryProviderStateMixin<PAL>));
+
+    return allocated;
+  }
+
+  private:
+
+    /**
+     * The last time we saw a low memory notification.
+     */
+    std::atomic<uint64_t> last_low_memory_epoch = 0;
+    std::atomic_flag lazy_decommit_guard = {};
+
 
     void new_block()
     {
@@ -71,11 +102,6 @@ namespace snmalloc
       remaining = SUPERSLAB_SIZE;
     }
 
-    /**
-     * The last time we saw a low memory notification.
-     */
-    std::atomic<uint64_t> last_low_memory_epoch = 0;
-    std::atomic_flag lazy_decommit_guard;
     SNMALLOC_SLOW_PATH void lazy_decommit()
     {
       // If another thread is try to do lazy decommit, let it continue.  If
@@ -141,11 +167,6 @@ namespace snmalloc
     }
 
   public:
-    /**
-     * Stack of large allocations that have been returned for reuse.
-     */
-    ModArray<NUM_LARGE_CLASSES, MPMCStack<Largeslab, PreZeroed>> large_stack;
-
     /**
      * Primitive allocator for structure that are required before
      * the allocator can be running.
@@ -422,5 +443,8 @@ namespace snmalloc
    * The memory provider that will be used if no other provider is explicitly
    * passed as an argument.
    */
-  inline GlobalVirtual default_memory_provider;
+  inline GlobalVirtual& default_memory_provider()
+  {
+    return *(Singleton<GlobalVirtual*, GlobalVirtual::make>::get());
+  }
 } // namespace snmalloc