Miscompilation of std::thread::scope on x86_64 with llvm sysroot

[bjorn3]

fn main() {
    std::thread::scope(|s| {
        s.spawn(|| {});
    });
}

$ ./y.sh build --sysroot llvm
$ dist/bin/rustc-clif foo.rs
$ RUST_BACKTRACE=1 ./foo

thread 'main' panicked at foo.rs:2:5:
a scoped thread panicked
stack backtrace:
   0: rust_begin_unwind
             at /rustc/65ea825f4021eaf77f1b25139969712d65b435a4/library/std/src/panicking.rs:619:5
   1: core::panicking::panic_fmt
             at /rustc/65ea825f4021eaf77f1b25139969712d65b435a4/library/core/src/panicking.rs:72:14
   2: std::thread::scoped::scope
   3: foo::main
   4: core::ops::function::FnOnce::call_once
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
Aborted (core dumped)

[bjorn3]

This blocks an update of the regex version we test.

[bjorn3]

Somewhat reduced version:

fn main() {
    std::thread::scope(|s| {
        std::thread::Builder::new().spawn_scoped(s, || {}).expect("failed to spawn thread");
    });
}

[bjorn3]

bytecodealliance/wasmtime#7210 seems to be the root cause of this. This issue doesn't show up when only using functions compiled with Cranelift as Cranelift doesn't ever take advantage of the uext and sext argument flags.