x86: use simd::intrinsics for bitshifts

[okaneco]

• Use intrinsics in sse2, avx2, avx512f, and avx512bw
• Mark intrinsics and tests that call these functions as const

Nothing notable about the implementation.

Intrinsics take count as the shift argument with the shift being a u64 in the low 64-bits of an __m128i. The shift amount must be less than the bit width of the type being shifted.

[rustbot]

r? @Amanieu

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

[Amanieu]

Can you check that LLVM is actually able to eliminate this check and fold it into the psllw instruction? It might still be keeping the check and only doing a psllw on the else branch.

[okaneco]

Good point, I assumed the instruction threshold tests were low enough to catch that.

But it looks like neither left nor right shifts fold into the instructions we want. Branches for the left shifts and cmov for the right shifts.
https://rust.godbolt.org/z/rzsehj5b8