RFR: 8348658: [AArch64] The node limit in compiler/codegen/TestMatcherClone.java is too strict
Vladimir Kozlov
kvn at openjdk.org
Wed Jan 29 18:41:50 UTC 2025
On Wed, 29 Jan 2025 15:33:43 GMT, Daniel Lundén <dlunden at openjdk.org> wrote:
> On some aarch64 machines, the node limit in the test `compiler/codegen/TestMatcherClone.java` is too strict.
>
> ### Changeset
>
> On the aarch64 machine I used when developing the fix for [JDK-8331295](https://bugs.openjdk.org/browse/JDK-8331295), the fix caused the node count in the test to decrease from 522 to 174. As a result, I set the IR framework node count limit to <200 to capture this. @shipilev now reports that on another aarch64 machine, the fix instead results in a decrease from 696 to 350. Since the fix for [JDK-8331295](https://bugs.openjdk.org/browse/JDK-8331295) still applies (the node count indeed decreases as expected), we should adjust the limit to somewhere in between 350 and 522 to capture both cases. 400 seems like a reasonable value and is what I propose in this PR.
>
> ### Testing
>
> - [GitHub Actions](https://github.com/dlunde/jdk/actions/runs/13014279125)
> - `tier1` and HotSpot parts of `tier2` and `tier3` on Windows x64, Linux x64, Linux aarch64, macOS x64, and macOS aarch64.
> - Extra testing of `TestMatcherClone.java` with various flag combinations.
Agree
-------------
Marked as reviewed by kvn (Reviewer).
PR Review: https://git.openjdk.org/jdk/pull/23349#pullrequestreview-2581906696
More information about the hotspot-compiler-dev
mailing list