add rtm cpu feature intrinsics

[mtak-]

Blocked by rust-lang/rust#60060

This does not include any HLE functions (xacquire/xrelease prefixed operations).

Let me know what needs changing.

[gnzlbg]

Sorry for the late review, this looks good to me.

[gnzlbg]

This LGTM. Can you ping me here when the upstream PR is merged? (or did that already happen?) I'll re-run CI then and merge it when its green.

[mtak-]

Thanks! It's merged, but not yet in nightly AFAICT. I'll ping you.

[mtak-]

Just changed this to return u8 to match intel headers more directly. It is specified to always be 0 or 1.

Additionally, if/when stdsimd later adds HLE intrinsics, _xtest should be enabled if either hle or rtm is supported.

[gnzlbg]

Additionally, if/when stdsimd later adds HLE intrinsics, _xtest should be enabled if either hle or rtm is supported.

I don't think this is something that can be easily supported. We need to know which features are available when generating code for this intrinsic, so it is either rtm, or hle, or both. But if its rtm or hle, then we need to generate different code depending on the answer to which one is it.

[mtak-]

All the intel docs say either HLE or RTM. The instruction set reference for xtest says under "CPUID Feature Flag": HLE or RTM.

we need to generate different code depending on the answer to which one is it.

Does that mean we need a duplicate of _xtest (_hletest) in any future hle.rs module?

@gnzlbg This should be ready to run through travis now.

[gnzlbg]

https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_xtest&expand=6168 says RTM CPUID flag only.

Does that mean we need a duplicate of _xtest (_hletest) in any future hle.rs module?

That's what it would mean, but that would be very weird. It seems more likely that the link you posted is incorrect, you might want to ask Intel for clarification in their forums.

[gnzlbg]

Can you rebase on top of master?

[mtak-]

I have already done so. verify signatures on master is broken on CI after #733.

---- verify_all_signatures stdout ----
failed to verify `_rdtsc`
  * failed to equate: `__int64` and PrimUnsigned(64) for _rdtsc
thread 'verify_all_signatures' panicked at 'assertion failed: all_valid', crates/stdsimd-verify/tests/x86-intel.rs:186:5

[gnzlbg]

Oh I thought the fix for that already had landed!

[gnzlbg]

Thank you @mtak- for working on this!