Fix mistake in #1928

[sayantn]

The x86 shift intrinsics do an overflowing shift (i.e. 0 in case of shift larger than type size), instead of wrapping shift (i.e. shift amount is taken modulo type size). So, we can't use simd_sh{l,r} here.

Also, simd_funnel_sh{l,r} doc says its UB for the shift amount to be greater than or equal to the type size. So, and them with T::BITS - 1 to ensure soundness (LLVM is clever enough to sees through it, there is no codegen impact)

[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]

wtf? random segfault??

edit: what has github been doing recently? Seeing so many spurious failures!

[madhav-madhusoodanan]

x86-64_pc-windows-msvc has been giving segfaults randomly, been noticing it too frequently when building - #1894

[sayantn]

Weird, but expected from msvc lol

Edit: this can possibly be something wrong on our side too, one test can be UBing and somehow only msvc is able to raise a segfault