[compiler-v2] v2 uses more stack space

[wrwg]

Given the following source:

//# print-bytecode --input=module
module 0xc0ffee::m {
   fun id_mut<T>(r: &mut T): &mut T {
        r
    }

    fun t0() {
        let x = &mut 0;
        let y = id_mut(x);
        *y;
        *x;
    }
}

... v1 creates bytecode which pops the unused *y immediately from the stack:

// Move bytecode v6
module c0ffee.m {


id_mut<Ty0>(Arg0: &mut Ty0): &mut Ty0 /* def_idx: 0 */ {
B0:
	0: MoveLoc[0](Arg0: &mut Ty0)
	1: Ret
}
t0() /* def_idx: 1 */ {
L0:	loc0: u64
L1:	loc1: &mut u64
B0:
	0: LdU64(0)
	1: StLoc[0](loc0: u64)
	2: MutBorrowLoc[0](loc0: u64)
	3: StLoc[1](loc1: &mut u64)
	4: CopyLoc[1](loc1: &mut u64)
	5: Call id_mut<u64>(&mut u64): &mut u64
	6: ReadRef
	7: Pop
	8: MoveLoc[1](loc1: &mut u64)
	9: ReadRef
	10: Pop
	11: Ret
}
}

But v2 waits with the pop until the end:

// Move bytecode v7
module c0ffee.m {


id_mut<Ty0>(Arg0: &mut Ty0): &mut Ty0 /* def_idx: 0 */ {
B0:
	0: MoveLoc[0](Arg0: &mut Ty0)
	1: Ret
}
t0() /* def_idx: 1 */ {
L0:	loc0: u64
L1:	loc1: &mut u64
L2:	loc2: &mut u64
B0:
	0: LdU64(0)
	1: StLoc[0](loc0: u64)
	2: MutBorrowLoc[0](loc0: u64)
	3: StLoc[1](loc1: &mut u64)
	4: CopyLoc[1](loc1: &mut u64)
	5: Call id_mut<u64>(&mut u64): &mut u64
	6: ReadRef
	7: MoveLoc[1](loc1: &mut u64)
	8: ReadRef
	9: Pop
	10: Pop
	11: Ret
}
}

Need to investigate whether this is a sign of a deeper issue in stack size usage in real world code.