JIT: Use unsigned native return types for small structs

[jakobbotsch]

Prefer the return type to be unsigned when small structs are returned in registers. This causes the backend to use zero extension instead of sign extension in a few cases, which has a smaller encoding on xarch.

Fixes a few small regressions I saw in #86388.

E.g.

[MethodImpl(MethodImplOptions.NoInlining)]
public static (byte, byte) Foo(in (byte, byte) arg) => arg;

Before:

G_M50947_IG02:  ;; offset=0000H
       movsx    rax, word  ptr [rcx]
						;; size=4 bbWeight=1 PerfScore 4.00

G_M50947_IG03:  ;; offset=0004H
       ret      

After:

G_M50947_IG02:  ;; offset=0000H
       movzx    rax, word  ptr [rcx]
						;; size=3 bbWeight=1 PerfScore 2.00

G_M50947_IG03:  ;; offset=0003H
       ret      

[ghost]

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

Issue Details

---

Prefer the return type to be unsigned when small structs are returned in registers. This causes the backend to use zero extension instead of sign extension in a few cases, which has a smaller encoding on xarch.

Fixes a few small regressions I saw in #86388.

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

[jakobbotsch]

cc @dotnet/jit-contrib PTAL @EgorBo.

Small number of (non-test) improvement. Few regressions related to CSE/register allocation.

[EgorBo]

This is a correctness change, right? Since, I guess, sign extension is not expected when you have C# byte

[jakobbotsch]

This is a correctness change, right? Since, I guess, sign extension is not expected when you have C# byte

No, we don't make any guarantees about normalization when structs are returned in registers, so there's no correctness issue with the existing codegen.