[jdk18] RFR: 8271202 C1: assert(false) failed: live_in set of first block must be empty

[Vladimir Ivanov]

On Fri, 24 Dec 2021 19:19:35 GMT, Igor Veresov <iveresov at openjdk.org> wrote:

> The root cause seems to be because of the irreducible loops (and therefore an unusual block traversal order when inserting phis) the phi invalidation logic in try_merge() doesn't invalidate phis that have invalid locals as inputs. I've attached a drawing: [8271202.pdf](https://github.com/openjdk/jdk18/files/7775050/8271202.pdf). Notice that i54 = phi (i43, 96) is not invalidated even though 96 is illegal. Transitively, i43, it should be illegal too. I would propose that we add a check for that and bailout in move_phi().
> 
> This has also been discussed here: https://github.com/openjdk/jdk/pull/6683
> 
> Testing is clean.

Looks good.

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

Marked as reviewed by vlivanov (Reviewer).

PR: https://git.openjdk.java.net/jdk18/pull/73