Whitelist pclmulqdq x86 feature flag

[newpavlov]

Relevant stdsimd issue.

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @michaelwoerister (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

[alexcrichton]

Thanks @newpavlov! From our discussion on rust-lang/stdarch#320 I think that we may want to rename this to pclmulqdq, though, as that's what Intel is calling it. This'll be the first of its kind doing that though so will require a little bit more effort unfortunately, but I think I can help guide you through what's necessary!

I think there may wish to be a few things we'd like to change as part of this:

• First, I think we'll want to store and pass features as &'static str rather than &CStr. This'll basically involve removing the \0 from the end of all the features.
• Next we'll want a function like target_feature_to_llvm_feature which requires a String. The mapping logic would then go in this function.
• You'll want to call target_feature_to_llvm_feature from two locations:
  • Here when we ask LLVM if a feature is enabled.
  • And here where we build up the attribute we'll attach to the LLVM function in the IR.

And... I think that's it! I think it's fine to define target_feature_to_llvm_feature in a "naive" fashion that looks like:

pub fn target_feature_to_llvm_feature(s: &str) -> String {
    match s {
        "pclmulqdq" => "pclmul".to_string(),
        s => s.to_string()
    }
}

and we can continue to make it more robust as needed

[retep998]

AMD also refers to it as PCLMULQDQ in their documentation.

[newpavlov]

@alexcrichton
Thank you!

First, I think we'll want to store and pass features as &'static str rather than &CStr. This'll basically involve removing the \0 from the end of all the features.

If I understand correctly null termination is needed for calling LLVMRustHasFeature. We could convert null-less &'static str features to CString, but it will require an allocation. Cost should be quite negligible and removing null bytes from features lists looks nice, but I can't shake off "allocations are evil" feeling. :) So I would like to hear your opinion on this.

[alexcrichton]

Ah yeah no need to worry about that, I can guarantee it won't show up in any profiles :)

[alexcrichton]

@bors: r+

Thanks!

[bors]

📌 Commit 7c6adb4 has been approved by alexcrichton

[ollie27]

This is a dangling pointer. See temporary_cstring_as_ptr.

[newpavlov]

Yes, thank you for noticing!

[alexcrichton]

@bors: r+

[bors]

📌 Commit c2a31de has been approved by alexcrichton

[gnzlbg]

Could we maybe just have a hashmap from rust feature name to llvm feature name, and if a feature is not in the hashmap, we just use the rust feature name as llvm's feature name ?

EDIT: Ah, I see you have a match statement, that looks good for me :)

[alexcrichton]

@bors: rollup