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

[Usha Gupta via llvm-commits]

================
@@ -2322,8 +2343,39 @@ PreservedAnalyses PostOrderFunctionAttrsPass::run(LazyCallGraph::SCC &C,
     Functions.push_back(&N.getFunction());
   }
 
-  auto ChangedFunctions =
-      deriveAttrsInPostOrder(Functions, AARGetter, ArgAttrsOnly);
+  bool NoFunctionsAddressIsTaken = false;
+  // Check if any function in the whole program has its address taken or has
+  // potentially external linkage.
+  // We use this information when inferring norecurse attribute: If there is
+  // no function whose address is taken and all functions have internal
+  // linkage, there is no path for a callback to any user function.
+  if (IsLTOPostLink || ForceLTOFuncAttrs) {
+    bool AnyFunctionsAddressIsTaken = false;
+    // Get the parent Module of the Function
+    Module &M = *C.begin()->getFunction().getParent();
+    for (Function &F : M) {
+      // We only care about functions defined in user program whose addresses
+      // escape, making them potential callback targets.
+      if (F.isDeclaration())
+        continue;
+
+      // If the function is already marked as norecurse, this should not block
+      // norecurse inference even though it may have external linkage.
+      // For ex: main() in C++.
+      if (F.doesNotRecurse())
+        continue;
----------------
usha1830 wrote:

@nikic I was further examining the behavior of this example, both with and without my patch.
In all the cases I've tested — including forcing `norecurse`, compiling as a shared library vs. an executable, using LTO and non-LTO options — the output remains the same, regardless of the patch.

However, it appears that  `ReversePostOrderPass` is adding the `norecurse` attribute to function `a`,  even when the `norecurse_fn` is **not** internal. 
The reason seems to be that all of `a`'s callers (`norecurse_fn()` and `foo()`) are themselves marked `norecurse`. Please note that  `norecurse` is inferred for `foo()` as well in this pass only.
This behavior doesn't seem entirely correct — or am I missing something?
https://godbolt.org/z/G8o9PczGG

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