RFR: 8215477: [lworld] C2 should avoid value type allocations when inlining incrementally
tobias.hartmann at oracle.com
Fri Feb 15 14:31:18 UTC 2019
On 15.02.19 13:48, Roland Westrelin wrote:
> Right. But is igvn.remove_dead_node(n) better/different than pushing on
> igvn worklist?
It's more explicit and saves another round of igvn processing of the dead node (and other nodes that
become dead as well).
>> Just wondering, why not always create a ValueTypePtrNode here?
> I can't see when that would help. Can you think of a scenario where
> having a ValueTypePtrNode as an input to a non inlined call would help?
Right, I can't think of such a scenario either because we never inline afterwards if
is_late_inline() is false.
More information about the valhalla-dev