Otherwise, the register allocator may spill and reload constants that
can be rematerialized with a single instruction.

These aren't AVX512VL tests

Resolves #161337

This test was failing with the internal shell due to the use of
subshells. This was not caught in my initial round of testing due to me
only using a M4 Mac for running my tests.

Extends the `icmp(trunc(shl))` fold to handle any power of 2 constant as
the shift base, not just 1. This generalizes the following patterns by
adjusting the comparison offsets by `log2(Pow2)`.

```llvm
(trunc (1 << Y) to iN) == 0    --> Y u>= N
(trunc (1 << Y) to iN) != 0    --> Y u<  N
(trunc (1 << Y) to iN) == 2**C --> Y ==  C
(trunc (1 << Y) to iN) != 2**C --> Y !=  C

; to

(trunc (Pow2 << Y) to iN) == 0    --> Y u>= N - log2(Pow2)
(trunc (Pow2 << Y) to iN) != 0    --> Y u<  N - log2(Pow2)
(trunc (Pow2 << Y) to iN) == 2**C --> Y ==  C - log2(Pow2)
(trunc (Pow2 << Y) to iN) != 2**C --> Y !=  C - log2(Pow2)
```

Proof: https://alive2.llvm.org/ce/z/2zwTkp

destroyLocals() does the same thing.

Allowing VPERMILPD/S intrinsics to be used in constexpr

Closes #167878

`[[nodiscard]]` should be applied to functions where discarding the
return value is most likely a correctness issue.
-
https://libcxx.llvm.org/CodingGuidelines.html#apply-nodiscard-where-relevant

It turns out that having `-ffast-math` as the only option to control
optimizations for MOD for REAL kinds (PR #160660) is too coarse-grained
for some applications. Thus, this PR adds back `-ffast-real-mod` to have
more control over the optimization. The `-ffast-math` flag will still
enable the optimization, and `-fno-fast-real-mod` allows one to disable
it.