[llvm] [FuncAttrs] Relax norecurse attribute inference (PR #139943)

[David Sherwood via llvm-commits]

================
@@ -0,0 +1,157 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --version 5
+; RUN: opt < %s -passes="lto<O2>" -S | FileCheck %s
+
+; This test includes a call graph with multiple SCCs. The purpose of this is
+; to check that norecurse is not added when a function is part of non-singular
+; SCC.
+; There are three different SCCs in this test:
+;  SCC#1:  main, foo, bar, foo1, bar1
+;  SCC#2:  bar2, bar3, bar4
+;  SCC#3:  baz, fun
+; None of these functions should be marked as norecurse
+
+; Function Attrs: nofree noinline nosync nounwind memory(none) uwtable
+define internal void @bar1() local_unnamed_addr #0 {
+; CHECK: Function Attrs: nofree noinline nosync nounwind memory(none) uwtable
+; CHECK-LABEL: define internal fastcc void @bar1(
+; CHECK-SAME: ) unnamed_addr #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 @main()
+; CHECK-NEXT:    ret void
+;
+entry:
+  %call = tail call i32 @main()
+  ret void
+}
+
+; Function Attrs: nofree noinline nosync nounwind memory(none) uwtable
+define dso_local noundef i32 @main() local_unnamed_addr #0 {
----------------
david-arm wrote:

I think it's worth adding `norecurse` to `main` here because you want `NoFunctionsAddressIsTaken` to be true in order to test your new code, right?

https://github.com/llvm/llvm-project/pull/139943