[Backport] 8333334: C2: Make result of Node::dominates more precise to enhance scalar replacement

[MaxXSoft]

[Backport] 8333334: C2: Make result of Node::dominates more precise to enhance scalar replacement

Summary: This patch changes the algorithm of Node::dominates to make the result more precise, allows the iterators of ConcurrentHashMap to be scalar replaced, and makes the iteration ~30% faster.

Testing: CI testing, IR test

Reviewers: kuaiwei, JoshuaZhuwj

Issue: #104

[MaxXSoft]

This backport is not clean due to missing backport of JDK-8287061 (openjdk/jdk#12897), which is too complex to backport.

Before applying this backport:

Benchmark                                  (nkeys)  Mode  Cnt        Score       Error  Units
Maps.testConcurrentHashMapIterators          10000  avgt   15   390877.321 ±  3723.284  ns/op

After:

Benchmark                                  (nkeys)  Mode  Cnt        Score       Error  Units
Maps.testConcurrentHashMapIterators          10000  avgt   15   330291.004 ± 14000.149  ns/op

[JoshuaZhuwj]

有个建议：因为是patch backport到dw21，所以我们需要考虑将来“后续patch backport”，以及“自动和jdk21同步”是否会产生冲突，所以理想情况是，尽量减少将来backport & auto merge可能产生冲突的可能性。所以可以考虑删除引入“DomResult”，只引入"将Load node加入igvn的worklist“这一个单独的change。但是因为相关的函数本身在后续可能产生冲突的可能性很小，所以这个backport，没有问题。approve. Thanks!

[kuaiwei]

使用all_controls_dominate和maybe_all_controls_dominate的场合有规则吗

[MaxXSoft]

@kuaiwei

使用all_controls_dominate和maybe_all_controls_dominate的场合有规则吗

all_controls_dominate 和添加这个 patch 之前原有的 all_controls_dominate 含义一致；maybe_all_controls_dominate 在遇到死代码时会返回更精确的结果，其余和 all_controls_dominate 一致。

maybe_all_controls_dominate 只在 LoadNode::split_through_phi 中使用，也就是这个 patch 的核心实现：如果在判断 dom 的时候遇到死代码，就等死代码被删掉之后重新判断一次。

其余所有情况都使用 all_controls_dominate。

[kuaiwei]

LGTM

[MaxXSoft]

Thanks for the review!