[llvm] [InstCombine] Fold usub_sat((sub nuw C1, A), C2) to usub_sat(C1 - C2, A) or 0 (PR #82280)

[via llvm-commits]

elhewaty wrote:

@nikic, Sorry I will update it now
>     * vector test
Do we need the vector test? 
>     * test with commuted sub like (sub nuw i32 %a, 12), proof: https://alive2.llvm.org/ce/z/kVXDgW

For this test I think the optimizer converts sub to add and then add a negative number, Is there is a way to match additions with over flow like this:

```
%res = call {i8, i1} @llvm.uadd.with.overflow.i8(i8 %c1, i8 %c2)
  %overflow = extractvalue {i8, i1} %res, 1
  %notoverflow = xor i1 %overflow, -1
  call void @llvm.assume(i1 %notoverflow)

  %add = sub nuw i8 %a, %c1
  %cond = call i8 @llvm.usub.sat.i8(i8 %add, i8 %c2)
  ret i8 %cond
```




https://github.com/llvm/llvm-project/pull/82280