`transmute::<u8, bool>()` generates bad code

[ChayimFriedman2]

pub fn bar(v: u8) -> bool {
    unsafe { std::mem::transmute(v) }
}

Generates (https://rust.godbolt.org/z/xzYdeffxq, note the redundant and):

mov     eax, edi
and     al, 1
ret

This is because we generate a redundant trunc/zext:

rust/compiler/rustc_codegen_ssa/src/mir/block.rs

Lines 1820 to 1824 in 1dda298

	// NOTE(eddyb) the `from_immediate` and `to_immediate_scalar`
	// conversions allow handling `bool`s the same as `u8`s.
	let src = bx.from_immediate(src.immediate());
	let src_as_dst = bx.bitcast(src, bx.backend_type(dst.layout));
	Immediate(bx.to_immediate_scalar(src_as_dst, dst_scalar)).store(bx, dst);

define zeroext i1 @_ZN7example3bar17h1b5c1a9c0782d25bE(i8 %v) unnamed_addr #0 {
  %0 = alloca i8, align 1
  %1 = trunc i8 %v to i1
  %2 = zext i1 %1 to i8
  store i8 %2, ptr %0, align 1
  %3 = load i8, ptr %0, align 1, !range !2, !noundef !3
  %4 = trunc i8 %3 to i1
  br label %bb1

bb1:                                              ; preds = %start
  ret i1 %4
}

[Rageking8]

@rustbot label +C-bug +T-compiler +A-codegen +I-slow

[deltragon]

This seems related to #96140, where the transmute went the other way.
The conclusion there was:

#96140 (comment)

Yea, the only actionable thing I can see here is to remove the extra asm instructions for perf reasons or to allow follow up optimizations to trigger.

I guess let's close this and if someone encounters a repro for more complex programs that have missing opts we can revisit

[ChayimFriedman2]

Yep, this is a dup.