Omr: What are the semantics of ILOpCode::isTreeTop()?
In the OMR compiler, isTreeTop is one of the several properties a node opcode can have. My understanding was that if isTreeTop() returns true, then the node must be anchored under a TR::TreeTop. However, as pointed out by #1553 and #1546, this does not seem to be the case.
So, what exactly does it mean for an opcode to have isTreeTop() return true?
Some related/follow-up questions:
- Should the ILValidator check for correct placement of TreeTop nodes?
- Should Tril implicitly anchor "non-TreeTop" nodes under a TreeTop node, unless it is already anchored or the user explicitly requests it to not be anchored?
@0xdaryl @mstoodle @vijaysun-omr @andrewcraik
Leonardo2718
All 2 comments
An alternative interpretation of the isTreeTop is all nodes that TR::TreeTops point to must have the property set.
I do think once we figure it out, probably do need to be using the validator to ensure it.
mgaudet
on 22 Aug 2017
@mgaudet's interpretation is right.
I believe isTreeTop must return true for any opcode that may appear at the top of the tree (node that TR::TreeTop points to). e.g. a store opcode would return true and while it is possible for a store opcode to be at the top of the tree, it is also possible for it to appear under a null check opcode.
vijaysun-omr
on 22 Aug 2017
Related issues
0xdaryl
路
5Comments
rwy0717
路
6Comments
hzongaro
路
4Comments
fjeremic
路
3Comments
0xdaryl
路
5Comments
Most helpful comment
@mgaudet's interpretation is right.
I believe isTreeTop must return true for any opcode that may appear at the top of the tree (node that TR::TreeTop points to). e.g. a store opcode would return true and while it is possible for a store opcode to be at the top of the tree, it is also possible for it to appear under a null check opcode.