[llvm] [InstCombine] Generalize zext(add X, -C) + C folding (PR #191723)

[via llvm-commits]

================
@@ -4535,3 +4535,18 @@ define <2 x i32> @ceil_div_vec_multi_use(<2 x i32> range(i32 0, 1000) %x) {
 declare void @use_i32(i32)
 declare void @use_vec(<2 x i32>)
 declare void @fake_func(i32)
+; Fold (add (zext (add X, -C)), C) -> (zext X) if X u>= C.
----------------
Maiowaa wrote:

Thanks for the suggestion!

I've added negative tests for:
- the case where C doesn't fit in the inner type (e.g., C=260)
- the mismatch case where outer C and inner -C don't correspond

Let me know if you'd like any further adjustments.

https://github.com/llvm/llvm-project/pull/191723