Allow direct allocation in old pointer space.

src/hydrogen.cc

Index: src/hydrogen.cc
diff --git a/src/hydrogen.cc b/src/hydrogen.cc
index 204ec20adb8c2d3ff358e37d008625dc5829a1fa..cf80e668b10c2c0ce14fe40e700132ea7aa1fbcb 100644
--- a/src/hydrogen.cc
+++ b/src/hydrogen.cc
@@ -1058,6 +1058,10 @@ HValue* HGraphBuilder::BuildAllocateElements(HContext* context,
   total_size->ClearFlag(HValue::kCanOverflow);
 
   HAllocate::Flags flags = HAllocate::CAN_ALLOCATE_IN_NEW_SPACE;
+  if (FLAG_pretenure_objects) {
+    flags = static_cast<HAllocate::Flags>(
+        flags | HAllocate::CAN_ALLOCATE_IN_OLD_POINTER_SPACE);

[danno, 2013/03/05 12:18:18]

I think this belongs inside HAllocate for now if the intent is to test old space
allocation, that way we catch all new allocations and test them as they get
added.

[Hannes Payer (out of office), 2013/03/11 17:16:31]

I think it is cleaner to add it to the callers of HAllocate.

+  }

[mvstanton, 2013/03/05 09:35:57]

This is a cool example of how you request in hydrogen an old space allocation.
This place is allocating a FixedArray, I think as part of a JSArray? I think it
might be good to also have the "parent" JSArray object be in the old space too.

Also, there is no intelligence behind the pretenuring, yet. Is this just to show
that if you turn on the flag the mechanism works?

In that case maybe build a little extra stuff. What about
1) A runtime function to ask:

SPACE GetObjectSpace(obj)

2) a unit test that uses this to show that the old space allocations work. You'd
turn on the new flag for that test, otherwise it'll continue being off by
default.

--> Soon enough there'll be all kinds of neat rules behind the pre-tenuring
decision, and the complexity of this simple unit test will grow nicely.

[Hannes Payer (out of office), 2013/03/11 17:16:31]

Done.

   if (IsFastDoubleElementsKind(kind)) {
     flags = static_cast<HAllocate::Flags>(
         flags | HAllocate::ALLOCATE_DOUBLE_ALIGNED);