Review duplication of loads in the Jit

[SingleAccretion]

See #75790 (comment).

We have a few places in the Jit where loads can potentially be duplicated, one especially high-risk area is address-exposed locals (everything that uses the "spill if complex, clone if local" pattern). See also #78713.

It would also be useful to think about whether we can debug-check this sort of thing.

category:cq
theme:addressing-modes

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

See #75790 (comment).

We have a few places in the Jit where loads can potentially be duplicated, one especially high-risk area is address-exposed locals (everything that uses the "spill if complex, clone if local" pattern). Redundant relop optimization is another known source.

It would also be useful to think about whether we can debug-check this sort of thing.

RLO reproduction (requires JitNoCSE=1):

[MethodImpl(MethodImplOptions.NoInlining)]
private static int Problem(int* p)
{
    var b = *p == 1;
    var c = b;
    if (b)
    {
        JitUse(c);
        return 2;
    }

    return 1;
}

[MethodImpl(MethodImplOptions.NoInlining)]
public static void JitUse<T>(T arg) { }

Author:	SingleAccretion
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[JulieLeeMSFT]

@markples PTAL.

[markples]

I'm going to mark this as priority:2. However, in practice that is going to mean (1) #78713 and (2) audit some "obvious" things (superficial look at copy/clone sites, audit "obvious" places (does regalloc rematerialize values). I don't have plans for figuring out a debug check. In general, this is a "do we have bugs" work item which is very open-ended, though of course it's a new requirement so it makes sense to give it some direct thought.