Stop-the-world compaction phase

The CL implements the core logic for per-page compaction, but in
a stop-the-world pause. Even though it's performed during a pause, it
handles every page independetly as the final goal is to have a
concurrent implementation.

This CL doesn't handle updating the native roots. Also, the black
allocations since the marking-phase pause are not handled yet.

Test: art/test/testrunner/testrunner.py
Bug: 160737021
Change-Id: Ib0be20663e0f9f76ee66a2a42180c4bd3579e41b

1 files changed, 4 insertions, 0 deletions

diff --git a/runtime/mirror/object_array.h b/runtime/mirror/object_array.h
index a20c86b82e..9a53708018 100644
--- a/runtime/mirror/object_array.h
+++ b/runtime/mirror/object_array.h
@@ -150,6 +150,10 @@ class MANAGED ObjectArray: public Array {
   // REQUIRES_SHARED(Locks::mutator_lock_).
   template<typename Visitor>
   void VisitReferences(const Visitor& visitor) NO_THREAD_SAFETY_ANALYSIS;
+  template<typename Visitor>
+  void VisitReferences(const Visitor& visitor,
+                       MemberOffset begin,
+                       MemberOffset end) NO_THREAD_SAFETY_ANALYSIS;
 
   friend class Object;  // For VisitReferences
   DISALLOW_IMPLICIT_CONSTRUCTORS(ObjectArray);