x86: Add disableAVX2/512 options and check XCR0 for OS support

[BradleyWood]

No description provided.

[BradleyWood]

jenkins build all

[BradleyWood]

I suspect this change will fix eclipse-openj9/openj9#19408 on windows server 2012.

[BradleyWood]

FYI, @0xdaryl @JamesKingdon @hzongaro

[0xdaryl]

Jenkins build all

[BradleyWood]

Builds don't run on draft PRs. I will undraft, but please don't merge until I can test on different system configurations.

[0xdaryl]

Added WIP to the title.

[BradleyWood]

Jenkins build all

[BradleyWood]

I found an issue with TR_DisableAVX in openJ9 without this PR. I am investigating.

[BradleyWood]

jenkins build all

[BradleyWood]

@0xdaryl Can you take another review of this PR. I expect it will fix eclipse-openj9/openj9#19408 and customer issue. Would be nice to get those fixed for 0.49, if possible.

[BradleyWood]

This PR shouldn't break J9. However, J9 doesn't respect when OMR disables features. eclipse-openj9/openj9#20691 will be required to use any of these options.

[BradleyWood]

jenkins build all

[0xdaryl]

@BradleyWood : in case it wasn't apparent, while you were away this PR was reverted because of Windows compile failures.

The issue was with the undefined compilation symbol in the OMR::X86::CPU::is_feature_disabled(uint32_t feature) function. Given that a CPU class is attached to every compilation object now, it makes sense to cache the compilation object in the CPU class if it exists for just these kinds of checks. Of course, a NULL compilation object is still a possibility if this class is used outside of the JIT or in early/late initialization. However, I'll let you consider the alternatives for how to resolve.