RFR: 8300638: Tier1 IR Test failure after JDK-8298632 on macosx-x64-debug

Pengfei Li pli at openjdk.org
Sat Jan 21 01:39:02 UTC 2023


On Sat, 21 Jan 2023 01:16:40 GMT, Vladimir Kozlov <kvn at openjdk.org> wrote:

> Which is strange.

Hi @vnkozlov , I'd like to explain more about this.

These jtreg tests are also used as (a kind of) fuzzer tests to ensure the correctness of C2 auto-vectorization. All annotated test methods are firstly forced compiled by C2 and then return values from executing C2 compiled code are compared with those from interpreted execution. The compilation control of C2 or interpreter is done with the WhiteBox API. But if we specify additional compiler control options, such as `-Xcomp` or `-XX:-TieredCompilation`, WhiteBox compiler control won't work any more. That's why we cannot simply remove `@requires vm.flagless`.

For now, I believe removing `@requires vm.flagless` is a real requirement. I will re-consider doing this but prefer trying it later after these x86 tier1 failures are fixed. Please let me know if you have any good ideas about it.

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

PR: https://git.openjdk.org/jdk/pull/12124


More information about the hotspot-compiler-dev mailing list