8388193: [lworld] C2's ClearArray keeps zero fill value alive in register

[TobiHartmann]

Valhalla extended ClearArrayNode with a value input to support nonzero initialization of flattened/null-restricted arrays. Ordinary array clears still supplied a ConL(0) and on x86 the matcher treated that value as an input constrained to RAX. This extended the live range causing increasing spills and a performance regression with the DaCapo23-biojava-large benchmark.

The zero-fill rules now match immL0, only use RAX as a local temporary and materialize zero in MacroAssembler::clear_mem like mainline code does. The fix also renames word_copy_only to requires_word_fill, improved comments and predicates and fixes some of the outdated opto assembly printing.

With this fix, we are back to baseline performance (from 47 ms/op to 35 ms/op).

Thanks,
Tobias

---

• 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 (1 review required, with at least 1 Committer)

Issue

• JDK-8388193: [lworld] C2's ClearArray keeps zero fill value alive in register (Bug - P4)

Reviewers

• Christian Hagedorn (@chhagedorn - Committer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/valhalla.git pull/2641/head:pull/2641
$ git checkout pull/2641

Update a local copy of the PR:
$ git checkout pull/2641
$ git pull https://git.openjdk.org/valhalla.git pull/2641/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 2641

View PR using the GUI difftool:
$ git pr show -t 2641

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/valhalla/pull/2641.diff

[bridgekeeper]

👋 Welcome back thartmann! A progress list of the required criteria for merging this PR into lworld 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.

[chhagedorn]

The fix looks good, thanks! And good clean-ups, I have some small comments following up on that.

[chhagedorn]

Should we also update xmm_clear_mem to xmm_fill_mem?

We should also do these comment updates in the header file.

[chhagedorn]

There are still some references in the .ad file to L_zero_64_bytes that should also be updated to L_fill_64_bytes.

[chhagedorn]

There are some other hardcoded uses of ymm0, for example in rep_stos_evex(). Do they also need to be updated to $tmp?

[TobiHartmann]

Thanks for your thorough review Christian! I did another pass over all the affected code. I renamed "zero" and "clear" to "fill" where appropriate, replaced hardcoded registers in the opto assembly printing by the correct placeholders and also fixed some of the other comments. Your comments should be all addressed now.

[chhagedorn]

Looks good, thanks for the update!

[TobiHartmann]

Thanks for the review Christian!