Use SIMD intrinsics whereever possible

[sayantn]

Also, remove uses of some deprecated LLVM intrinsics

FIXME: _mm256_test{z,c}_{ps,pd} doesn't optimize well in the testsuite for some reason, it works on godbolt 😕

[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

[sayantn]

macos-15 runners are failing for some reason, is this a GH issue?

[Amanieu]

I restarted the runners and it seems to be working fine now. Do you want this merged now or try to find a solution for _mm256_test{z,c}_{ps,pd}?

[sayantn]

let's merge this for now. That can be another PR later (the testnzc intrinsics will also need some attention)

[sayantn]

Just discovered that the s{l, r}l{,v} intrinsics are wrongly implemented here, they should zero on shift amount overflow (like unbounded_shl), but do wrap on overflow (like wrapping_shl). I will send a pr to revert those

Edit: weirdly, the rotate instructions do wrap on overflow

[RalfJung]

Nice, this means we can remove the implementation of vperm2 from Miri. :)
Are there any similar recent patches / more implementations we can remove now?

[sayantn]

There should be quite a few, we have removed uses of many llvm intrinsics recently