Missed optimization combining saturating-add with min

[jrose-signal]

The following IR (simplified from optimized output of rustc)

; min(x.saturating_add(1), 5)
define noundef i32 @test(i32 noundef %x) {
start:
  %0 = tail call i32 @llvm.uadd.sat.i32(i32 %x, i32 1)
  %1 = tail call noundef i32 @llvm.umin.i32(i32 %0, i32 5)
  ret i32 %1
}

produces the following assembly

test:
	incl	%edi
	movl	$-1, %ecx
	cmovnel	%edi, %ecx
	cmpl	$5, %ecx
	movl	$5, %eax
	cmovbl	%ecx, %eax
	retq

but it could just be

test:
	cmpl	$5, %edi
	movl	$4, %eax
	cmovbl	%edi, %eax
	incl	%eax
	retq

or similar.

(Very similar to #92433, but with a constant as an argument to min instead of another addition or variable.)

[dtcxzyw]

Does this pattern exist in some real-world code?

[jrose-signal]

https://github.com/signalapp/libsignal/blob/f2ae8436b365f5e4e1371102f4702f51ac34e02c/rust/net/src/infra/connection_manager.rs#L110-L115

Certainly we didn't have to write it this way (given the invariants, a non-saturating add would have been fine), but I don't think it's unreasonable.