Unnecessary compares and branches in asm output

[Boscop]

https://godbolt.org/g/62GNZf

pub fn f1(v: &mut Vec<f32>) {
  v.clear();
}

pub fn f2(v: &mut Vec<f32>) {
  unsafe { v.set_len(0); }
}

example::f1:
  push rbp
  mov rbp, rsp
  cmp qword ptr [rdi + 16], 0
  je .LBB0_2
  mov qword ptr [rdi + 16], 0
.LBB0_2:
  pop rbp
  ret

example::f2:
  push rbp
  mov rbp, rsp
  mov qword ptr [rdi + 16], 0
  pop rbp
  ret

In f1, the lines cmp qword ptr [rdi + 16], 0 and je .LBB0_2 should be optimized away, but aren't, why?
(It sets the len to 0 if it's not 0, otherwise leaves it at 0. The len will be 0 afterwards either way, so it should eliminate the check and just set it to 0 like set_len(0).)

I thought I'd open this issue because I don't think such optimization opportunities should be left on the table because it accumulatively affects a lot of code.

[asquared31415]

As of rust 1.32.0 onwards, f2 does not exist in the assembly output (I believe because f2 and f1 produce the same assembly) and f1 is optimized completely to just

mov qword ptr [rdi + 16], 0
ret

however most uses will get inlined anyway if possible.

[veera-sivarajan]

fixed since rustc 1.30.0: https://rust.godbolt.org/z/7qK1rojYG

@rustbot label +e-needs-test

[zakrad]

@rustbot claim