r? @wesleywiser

(rustbot has picked a reviewer for you, use r? to override)

⚠️ Warning ⚠️

• These commits modify submodules.

Cc @alexcrichton.

This won't currently compile until rust-lang/stdarch#1494 is merged and updated, so for now this is waiting for the FCP first.

Personally I'd say this is good to go (modulo CI of course), and it's something I forgot to do after the last CG meeting, thanks @daxpedda!

For others this PR has relevant discussion namely some of the background laid you in this comment. The relaxed-simd proposal is a bit different where there's actual instructions to expose with new intrinsics, but this is following in the footsteps of the simd128 proposal and its intrinsics so AFAIK there's no new precedents set here or anything like that, mostly just following preexisting processes.

@wesleywiser AFAIU this should probably not be marked as blocked.

It's true that merging is blocked on rust-lang/stdarch#1494, but the FCP is required for rust-lang/stdarch#1494 to be merged in the first place.
It could be argued that this is blocked on #117372, which in turn is blocked by tkaitchuck/aHash#183, but that should probably not block an FCP.

So I would like to request an FCP for stabilizing relaxed SIMD!
Let me know if I misunderstood something.

This cover the relaxed-simd target feature (lang team) and the related intrinsics in stdarch (libs-api team).

@rfcbot fcp merge

Team member @Amanieu has proposed to merge this. The next step is review by the rest of the tagged team members:

• @Amanieu
• @BurntSushi
• @dtolnay
• @joshtriplett
• @m-ou-se
• @nikomatsakis
• @pnkfelix
• @scottmcm
• @tmandry

No concerns currently listed.

Once a majority of reviewers approve (and at most 2 approvals are outstanding), this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

cc @rust-lang/lang-advisors: FCP proposed for lang, please feel free to register concerns.
See this document for info about what commands tagged team members can give me.

Repeating what I posed on the tracking issue:

I am confused by the spec here. Quoting from https://github.com/WebAssembly/relaxed-simd/blob/main/proposals/relaxed-simd/Overview.md:

The same instruction at (1) and (2), when given the same inputs, can return two different results.

Okay, so that's normal non-determinism then. Wasm already has that for NaNs when there are non-canonical inputs, which is strangely not acknowledged here (has that changed?), but sure, makes sense.

Some operators are host-dependent, because the set of possible results may depend on properties of the host environment (such as hardware). Technically, each such operator produces a fixed-size list of sets of allowed values. For each execution of the operator in the same environment, only values from the set at the same position in the list are returned, i.e., each environment globally chooses a fixed projection for each operator.

This describes something else! IIUC, it says that the operation, when executed twice in the same environment, must produce the same result. In other words, the non-determinism is fixed once at program startup, and the example givejn just above actually cannot happen within a single program/module execution. This is closer to what we usually call implementation-specific behavior than to full non-determinism.

Can someone explain this seeming contradiction?

To the best of my knowledge I believe that the contradiction is explained by the fact that the proposal has a long history and changed at one point. I believe your first quoted sentence is the original semantics when the proposal was first drafted and the second quote you listed is the desired semantics.

For example this was the commit that introduced the first quote and it refers to an fpenv idea which was later removed. While I can't speak with complete certainty I'm pretty confident you've discovered a mistake in the overview. This could be confirmed with the rendered specification but that is a broken link at this time. I'm not great at reading the raw spec myself.

Okay, I've opened WebAssembly/relaxed-simd#153 to hopefully get clarification from the authors.

For Rust, we should probably conservatively treat these as being non-deterministic at each operation. I don't think that should cause any issues? These are just LLVM intrinsics implemented via certain wasm instructions, right?

I don't think that should cause any issues?

No, we do nothing with the output (the part that might not be deterministic).

These are just LLVM intrinsics implemented via certain wasm instructions that, right?

Indeed.

☔ The latest upstream changes (presumably #117915) made this pull request unmergeable. Please resolve the merge conflicts.

@bors: r

📌 Commit 61d95e2 has been approved by alexcrichton

It is now in the queue for this repository.

It looks like this may have caused the failure here #128454 (comment). Is it possible to update the stdarch submodule separately from stabilizing the wasm feature? That might be better in its own PR (and at least its own commit, for better history).

@bors r-

Sure!
Will do that as soon as I figure out what the problem is exactly.

@Amanieu these errors look like AVX512 work related to the Stdarch update, would you know whats causing this or who to ping?

It could also just be the RFL job having something wrong with target features, I brought it up on Zulip too.

Stdarch update is happening in #128466, which this PR will be rebased on after its merged.

#128466 (comment) merged so this should be ready for a rebase.

@bors: r

📌 Commit 80b74d3 has been approved by alexcrichton

It is now in the queue for this repository.

⌛ Testing commit 80b74d3 with merge 9179d9b...

☀️ Test successful - checks-actions
Approved by: alexcrichton
Pushing 9179d9b to master...

Finished benchmarking commit (9179d9b): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 757.802s -> 757.898s (0.01%)
Artifact size: 336.77 MiB -> 336.81 MiB (0.01%)