Revert "Use SIMD intrinsics for vector shifts"

[RalfJung]

Reverts #1955 to fix rust-lang/rust#159831 and to fix rust-lang/rust#159801.
Cc @sayantn

No tests because the per-intrinsic codegen is actually fine, it's the combined codegen with other operations where things fall apart.

[rustbot]

r? @adamgemmell

rustbot has assigned @adamgemmell.
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: @Amanieu, @adamgemmell, @davidtwco, @folkertdev, @sayantn
• @Amanieu, @adamgemmell, @davidtwco, @folkertdev, @sayantn expanded to Amanieu, adamgemmell, davidtwco, folkertdev, sayantn
• Random selection from Amanieu, adamgemmell, davidtwco, folkertdev, sayantn

[RalfJung]

(CI seems to be generally broken, that's not from my PR.)

[adamgemmell]

CI should be fixed if you give this a rebase

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[RalfJung]

#7 0.384 curl: (23) Failure writing output to destination, passed 1494 returned 18

CI seems broken in new ways now.

[RalfJung]

I don't know what's going on with that Windows job. The 691 failures seem to all be similar:

thread 'core_arch::x86::avx512fp16::tests::test_mm256_castpd_ph' (1328) panicked at crates\core_arch\src\x86\avx512fp16.rs:17098:9:
assertion `left == right` failed
  left: Simd(0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00)
 right: Simd(0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x3c00)

Those are the same? Also the PR doesn't touch that intrinsic...?

[RalfJung]

Wtf, it worked on a retry...

[adamgemmell]

I couldn't find the failing windows job in the gha history, do you happen to have a link to it?

View changes since this review

[RalfJung]

I couldn't find the failing windows job in the gha history, do you happen to have a link to it?

https://github.com/rust-lang/stdarch/actions/runs/34347799442/job/102462592457?pr=2217

[adamgemmell]

I couldn't find the failing windows job in the gha history, do you happen to have a link to it?

https://github.com/rust-lang/stdarch/actions/runs/34347799442/job/102462592457?pr=2217

Spooky. Possibly it's due to fp16 just being broken on windows gnu, see llvm_util.rs and https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115054. On a dev profile where inlining doesn't happen maybe the equality check can fail while the debug print works fine. Perhaps we should skip those tests on windows gnu.