Fix safety doc in intrinsics::simd

[yilin0518]

The Safety documentation for simd_gather and simd_scatter incorrectly described unmasked values in T as requiring readable or writable memory.

The memory access is performed through the corresponding enabled pointer lane in ptr, so this updates the documentation to state that requirement explicitly.

[rustbot]

Some changes occurred to the intrinsics. Make sure the CTFE / Miri interpreter
gets adapted for the changes, if necessary.

cc @rust-lang/miri, @RalfJung, @oli-obk, @lcnr

Some changes occurred to the platform-builtins intrinsics. Make sure the
LLVM backend as well as portable-simd gets adapted for the changes.

cc @antoyo, @GuillaumeGomez, @bjorn3, @calebzulawski, @programmerjake

[rustbot]

r? @JohnTitor

rustbot has assigned @JohnTitor.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: libs
• libs expanded to 12 candidates
• Random selection from 6 candidates

[RalfJung]

Is there LLVM documentation confirming the exact requirements here?

[RalfJung]

Looking at the diff, I think the old and new wording are equivalent. An "unmasked value" is likely meant to be interpreted as a value in a lane that has not been "masked out", i.e. in an enabled lane. But I agree that the old wording is confusing.

So r=me on the change, but CI failed because the doc link is not right.
@rustbot author
r? me

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[yilin0518]

Is there LLVM documentation confirming the exact requirements here?

I found the possible corresponding documentation: llvm-masked-gather-intrinsics and llvm-masked-scatter-intrinsics, in which it emphasizes that ptr should be readable/writable if the value in mask is !0.

[yilin0518]

@RalfJung I commit the fixed commit, it should have solved the link problem.

Besides, I open a topic in Zulip: Maybe some safety doc need fixs in intrinsics::simd. In this topic I refered another functions: simd_masked_load and simd_masked_store. I think the safety section of these function should add an extra entry: Enabled ptr elements must be safe to read/write as if by core::ptr::read/write. Can you provide any suggestion? Thank you!

Also, I think simd_funnel_shl, simd_funnel_shr, simd_shl, simd_shr need to be modified: change less than <int>::BITS to in 0..<int>::BITS. If really need, I can provide a new PR to fix these functions.

[RalfJung]

Now the crate:: will show up in the rendered text, that's not great.
Something like [`ptr::read`][crate::ptr::read] should work.

View changes since the review

[RalfJung]

You dud not copy what I suggested and did not fix the problem I think.

[yilin0518]

Oops, thank you for your remind. I think I do the right thing this time.😭

[RalfJung]

Also note that this is still marked as waiting-for-author. As the bot says, please say @rustbot ready if you want another review. Sometimes reviewers will guess that the PR is ready for review but that doesn't always work.

[yilin0518]

@rustbot ready

[RalfJung]

@rustbot author

[yilin0518]

@rustbot ready