require simd_insert, simd_extract indices to be constants

[RalfJung]

As discussed in #77477 (see in particular here). This PR doesn't touch codegen yet -- the first step is to ensure that the indices are always constants; the second step is to then make use of this fact in backends.

Blocked on rust-lang/stdarch#1530 propagating to the rustc repo.

[rustbot]

r? @oli-obk

rustbot has assigned @oli-obk.
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

[oli-obk]

I should investigate whether I can do the same trick as #115933 for insert and extract

[RalfJung]

Which trick?
Const-generic versions of simd_extract / simd_insert are definitely possible. But with the current status of const generics, stdarch couldn't use them, just like it can't use simd_shuffle_generic. I asked @lcnr about this recently but can't find the link right now...

[oli-obk]

Ah, they wanna do math on their own const generics. Yea, that won't work

[RalfJung]

I think all they want to do is as casts.^^ (Though I didn't do an exhaustive check to see if that covers everything. And for simd_shuffle, they also need to construct a slice.)

[oli-obk]

r=me when stdsimd has been synced

[RalfJung]

That should be soon hopefully: #121185

[RalfJung]

We had a codegen that that consisted only of calls to simd_insert / simd_extract with variable arguments for the index. @workingjubilee @Amanieu @calebzulawski what are we supposed to do with that, given that we said we want these functions to be like simd_shuffle and only support constant indices? I don't know what the test was actually supposed to test, it had "generic" in the name. See the last commit of this PR.

[RalfJung]

Looks like the test was added in #78863, Cc @KodrAus @gnzlbg

[calebzulawski]

I think changing those arguments to const generics is appropriate. Typeck or similar should probably have rejected the non-const argument (which is what shuffles do, I think?) It seems packed_simd is really the only consumer of those intrinsics, if std::arch doesn't use them.

[RalfJung]

stdarch uses them as well, but since rust-lang/stdarch#1530 it makes all arguments constants. It also had one non-constant use in a test only, somewhat similar to this codegen test.

But if we don't want to support non-constant simd_insert/extract, it makes little sense to have a codegen test for it.

Note that this does not quite make the arguments const generics, instead it makes them special magic arguments that must be const. Moving to const generics is tracked at #85229, but not currently possible yet.

[Amanieu]

I also think it's fine to make these require constants.

[RalfJung]

Is there something that I should do with that codegen test, or should I delete it?

[Amanieu]

It's fine to just remove it. We're changing the definition of simd_insert/simd_extract to only accept constants, which is fine since this is an unstable internal API. That test is no longer relevant.

[RalfJung]

Okay, thanks. :)
@bors r=oli-obk,Amanieu

[bors]

📌 Commit e19f89b has been approved by oli-obk,Amanieu

It is now in the queue for this repository.

[bors]

⌛ Testing commit e19f89b with merge 52dba5f...

[bors]

☀️ Test successful - checks-actions
Approved by: oli-obk,Amanieu
Pushing 52dba5f to master...

[rust-timer]

Finished benchmarking commit (52dba5f): comparison URL.

Overall result: ❌ regressions - no action needed

@rustbot label: -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.5%	[1.5%, 1.5%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	3.1%	[3.1%, 3.1%]	1
Regressions ❌ (secondary)	3.5%	[3.5%, 3.5%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	3.1%	[3.1%, 3.1%]	1

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: 649.949s -> 649.986s (0.01%)
Artifact size: 310.94 MiB -> 310.96 MiB (0.01%)