Add prefetch to mpscq.

src/ds/bits.h

@@ -136,6 +136,15 @@ namespace snmalloc
 #endif
     }
 
+    inline void prefetch(void* ptr)
+    {
+#if defined(PLATFORM_IS_X86)
+      _mm_prefetch(reinterpret_cast<const char *>(ptr), _MM_HINT_T0);
+#else
+#  warning "Missing pause intrinsic"
+#endif
+    }
+
     inline uint64_t tick()
     {
 #if defined(PLATFORM_IS_X86)

src/ds/mpscq.h

@@ -71,6 +71,7 @@ namespace snmalloc
       if (next != nullptr)
       {
         front = next;
+        bits::prefetch(&(next->next));
         assert(front);
         std::atomic_thread_fence(std::memory_order_acquire);
         invariant();