[llvm] [DAG] Fold (umin (sub a b) a) -> (usubo a b); (select usubo.1 a usubo.0) (PR #161651)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,14 @@
+; RUN: llc < %s -mtriple=aarch64 | FileCheck %s
+
+; GitHub issue #161036
+
+define i64 @underflow_compare_fold(i64 %a, i64 %b) {
+; CHECK-LABEL: underflow_compare_fold
+; CHECK:      // %bb.0:
+; CHECK-NEXT: subs x8, x0, x1
+; CHECK-NEXT: csel x0, x0, x8, lo
+; CHECK-NEXT: ret
+  %sub = sub i64 %a, %b
+  %cond = tail call i64 @llvm.umin.i64(i64 %sub, i64 %a)
+  ret i64 %cond
+}
----------------
arsenm wrote:

The point isn't whether it's correct or not, but whether it's worthwhile. If you aren't eliminating the instruction, is it worth doing the fold (I suppose maybe if the min will be expanded) 

https://github.com/llvm/llvm-project/pull/161651