RFR: 8274130: C2: MulNode::Ideal chained transformations may act on wrong nodes [v2]

[Aleksey Shipilev]

On Mon, 27 Sep 2021 08:58:26 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

>> I was puzzled by it when fixing JDK-8274060. It looks that new optimizations added by [JDK-8273454](https://bugs.openjdk.java.net/browse/JDK-8273454) and [JDK-8263006](https://bugs.openjdk.java.net/browse/JDK-8263006) rewire `in(1)` and `in(2)` in `MulNode::Ideal`, which means the chained transformations should see them? Yet, both inputs and their `Type`-s are cached locally and not refreshed. I have not seen failures due to this yet, but it looks that the current code is subtly incorrect because of this. 
>> 
>> I thought about doing `return this` instead of `progress = true`, so that we leave `MulNode::Ideal` once we hit any transform and hope to return back, but I wondered if that would expose us to different graph shapes in-between successive `MulNode::Ideal` calls, which might have other unintended consequences. Therefore, I opted to a more conservative patch.
>> 
>> Additional testing:
>>  - [x] `compiler/` tests
>>  - [x] `tier1` tests
>>  - [x] 100K Fuzzer tests (one unrelated failure)
>
> Aleksey Shipilev has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Align the stars manually

Thanks!

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

PR: https://git.openjdk.java.net/jdk/pull/5631