Codegen/allocator regression from 1.60

[jgouly]

In 1.60 the following code causes no allocations:

pub fn main() -> i32 {
    let v1 = vec![5, 6, 7];
    let v1_iter = v1.iter();
    let total: i32 = v1_iter.sum();
    println!("{}",total);
    total
}

However in nightly there is an allocation, see: https://godbolt.org/z/cq9b1Pnva

EDIT:

Appears there is also a problem with 1.60 and opt-level 2 vs 3: https://godbolt.org/z/8MWjEvPbh

[nox]

This feels like an optim pass in LLVM was promoted from level 3 to level 2. I checked on other archs and the problem also appears.

[nox]

@lqd suspects it's related to #94402. I checked 2022-03-01 and there is no alloc, while on 2022-03-02 the alloc comes back.

[nox]

The allocation also goes away with opt-level 3 if you pass -C panic=abort

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-high

[pnkfelix]

T-compiler P-high review, adding T-compiler label.

@rustbot label +T-compiler +A-LLVM

[pnkfelix]

T-compiler P-high review, downgrading to P-medium.

@rustbot label: -P-high +P-medium