Allocation sampling for paged/lo spaces

src/heap/spaces-inl.h

Index: src/heap/spaces-inl.h
diff --git a/src/heap/spaces-inl.h b/src/heap/spaces-inl.h
index 3023fbf51ead410fba6a14aaab99dff05e5acf61..55eed7dbb9eadaba463c27ed5bdb45fe6cb1c3ae 100644
--- a/src/heap/spaces-inl.h
+++ b/src/heap/spaces-inl.h
@@ -425,12 +425,24 @@ AllocationResult PagedSpace::AllocateRawAligned(int size_in_bytes,
 AllocationResult PagedSpace::AllocateRaw(int size_in_bytes,
                                          AllocationAlignment alignment) {
 #ifdef V8_HOST_ARCH_32_BIT
-  return alignment == kDoubleAligned
-             ? AllocateRawAligned(size_in_bytes, kDoubleAligned)
-             : AllocateRawUnaligned(size_in_bytes);
+  AllocationResult result =
+      alignment == kDoubleAligned
+          ? AllocateRawAligned(size_in_bytes, kDoubleAligned)
+          : AllocateRawUnaligned(size_in_bytes);
 #else
-  return AllocateRawUnaligned(size_in_bytes);
+  AllocationResult result = AllocateRawUnaligned(size_in_bytes);
 #endif
+  if (!result.IsRetry()) {
+    Object* obj = result.ToObjectChecked();
+    if (obj->IsHeapObject()) {

[ofrobots, 2016/01/23 16:16:31]

Why is this check necessary here?

[ofrobots, 2016/01/26 00:38:25]

*bump*

[mattloring, 2016/01/26 00:42:48]

An Object can be an Smi or a HeapObject so I check that here. Is the check known
to be unnecessary in this instance?

[ofrobots, 2016/01/26 15:47:15]

We just allocated an object on the heap; so this check should not be necessary I
think. 

+      HeapObject* heap_obj = static_cast<HeapObject*>(obj);
+      for (int i = 0; i < allocation_observers_.length(); ++i) {
+        AllocationObserver* o = allocation_observers_[i];
+        o->AllocationStep(size_in_bytes, heap_obj->address(), size_in_bytes);
+      }
+    }
+  }
+  return result;
 }