8381872: C2: assert before block local scheduling failed with UseCompactObjectHeaders

[rkennke]

This fixes a bug in Peephole::lea_remove_redundant() in peephole_x86_64.cpp, which can lead to miscompilations caused by two distinct values getting assigned to the same register.

Testing:

• tier1 -UCOH
• tier1 +UCOH

---

• I confirm that I make this contribution in accordance with the OpenJDK Interim AI Policy.

---

Progress

• Change must not contain extraneous whitespace
• Commit message must refer to an issue
• Change must be properly reviewed (2 reviews required, with at least 1 Reviewer, 1 Author)

Issue

• JDK-8381872: C2: assert before block local scheduling failed with UseCompactObjectHeaders (Bug - P3)

Reviewers

• Manuel Hässig (@mhaessig - Committer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/30762/head:pull/30762
$ git checkout pull/30762

Update a local copy of the PR:
$ git checkout pull/30762
$ git pull https://git.openjdk.org/jdk.git pull/30762/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 30762

View PR using the GUI difftool:
$ git pr show -t 30762

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/30762.diff

Using Webrev

Link to Webrev Comment

[bridgekeeper]

👋 Welcome back rkennke! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

[openjdk]

❗ This change is not yet ready to be integrated.
See the Progress checklist in the description for automated requirements.

[openjdk]

@rkennke The following label will be automatically applied to this pull request:

• hotspot-compiler

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

[mlbridge]

Webrevs

• 00: Full (925e9f4f)

[mhaessig]

Thank you for fixing this, @rkennke. The fix looks good to me.

I'm giving this a spin on our CI on higher tiers and will report back with the results.

[openjdk]

The total number of required reviews for this PR has been set to 2 based on the presence of this label: hotspot-compiler. This can be overridden with the /reviewers command.

[mhaessig]

I tested up to tier4 for for -UCOH and up to tier6 for +UCOH and everything passed.

[merykitty]

Is it necessary that the base input of a leaP... is non-null? Given that the leaP... does not appear in an oop map, can we make the base input nullptr, so we don't need to worry about register conflict anymore?

[dean-long]

I experimented with this so I could understand it better, and something seems to be off. The above code disables the optimization even if the only dependant_lea is the lone lea_derived_oop, in which case it would seem to be OK because we aren't replacing the base with a "new base".
Also, I tried disabling the above change, and the test below still passes.

[mhaessig]

Is it necessary that the base input of a leaP... is non-null? Given that the leaP... does not appear in an oop map, can we make the base input nullptr, so we don't need to worry about register conflict anymore?

You are absolutely right, @merykitty. The AddPNode::Base is only for the oop map and does not even get matched. This peephole is for cleaning up decodes that nevver get used in an oop map so rewiring is not even necessary. I wish I had thought of this a year ago...

The following does the trick (inspired by AddPNode::make_off_heap since using nullptr does not work):

Suggested change

	dependant_lea->set_req(AddPNode::Base, new_base);
	dependant_lea->set_req(AddPNode::Base, Compile::current()->top());

By not introducing a dependency, there will also not be a conflict.