[lworld] RFR: 8367242: [lworld] C2 compilation asserts with "dead loop detected"

[Tobias Hartmann]

On Tue, 30 Sep 2025 08:03:27 GMT, Marc Chevalier <mchevalier at openjdk.org> wrote:

>> src/hotspot/share/opto/inlinetypenode.cpp line 171:
>> 
>>> 169: 
>>> 170: // Merges 'this' with 'top' by updating the input PhiNodes added by 'clone_with_phis'
>>> 171: InlineTypeNode* InlineTypeNode::merge_with_top(PhaseGVN* gvn, int pnum, bool transform) {
>> 
>> I don't like the code duplication with `InlineTypeNode::merge_with` here. Would it be feasible to merge the two and have `InlineTypeNode::merge_with` handle `top` for `other`? Maybe the `PhiNode` inputs and `val2`can be initialized to `top` by default (maybe already in `clone_with_phis`) such that we only need to guard the `set_req` calls by `!other->is_top()`?
>
> If we initialize the inputs of the `PhiNodes` above `this` to top, we don't even need to merge with anything. We can just skip the call to `merge_with`, no? It seems mostly harmless to change `clone_with_phis` this way, except for the callsite in `Parse::ensure_phi` where I'm not sure of the consequences.

Right, skipping the call would be even better. I think the `ensure_phi` callsite should be fine as well as all the branches that the phi inputs correspond to should either be initialized afterwards or are dead (in which case `top` seems appropriate). But only testing will tell :D

-------------

PR Review Comment: https://git.openjdk.org/valhalla/pull/1640#discussion_r2390314232