Add multithreading test for SamplingCircularQueue, fix implementation.

src/globals.h

Index: src/globals.h
diff --git a/src/globals.h b/src/globals.h
index cb7f27ee7d235cfc2493e5ac85fe6ca4bb354ea3..90007e67bfb345a745905bd787e5ab5d12460d8d 100644
--- a/src/globals.h
+++ b/src/globals.h
@@ -195,6 +195,10 @@ const Address kFromSpaceZapValue = reinterpret_cast<Address>(0xbeefdad);
 // gives 8K bytes per page.
 const int kPageSizeBits = 13;
 
+// On Intel architecture, cache line size is 64 bytes.
+// On ARM it may be less (32 bytes), but as far this constant is
+// used for aligning data, it doesn't hurt to align on a greater value.
+const int kProcessorCacheLineSize = 64;
 
 // Constants relevant to double precision floating point numbers.