[llvm] Reapply [AMDGPU] Avoid resource propagation for recursion through multiple functions (PR #112251)

[Matt Arsenault via llvm-commits]

================
@@ -101,25 +101,28 @@ void MCResourceInfo::assignResourceInfoExpr(
   const MCConstantExpr *LocalConstExpr =
       MCConstantExpr::create(LocalValue, OutContext);
   const MCExpr *SymVal = LocalConstExpr;
+  MCSymbol *Sym = getSymbol(FnSym->getName(), RIK, OutContext);
   if (!Callees.empty()) {
     SmallVector<const MCExpr *, 8> ArgExprs;
-    // Avoid recursive symbol assignment.
     SmallPtrSet<const Function *, 8> Seen;
     ArgExprs.push_back(LocalConstExpr);
-    const Function &F = MF.getFunction();
-    Seen.insert(&F);
 
     for (const Function *Callee : Callees) {
       if (!Seen.insert(Callee).second)
         continue;
+
       MCSymbol *CalleeFnSym = TM.getSymbol(&Callee->getFunction());
       MCSymbol *CalleeValSym =
           getSymbol(CalleeFnSym->getName(), RIK, OutContext);
-      ArgExprs.push_back(MCSymbolRefExpr::create(CalleeValSym, OutContext));
+      if (!CalleeValSym->isVariable() ||
+          !CalleeValSym->getVariableValue(/*isUsed=*/false)
+               ->isSymbolUsedInExpression(Sym)) {
----------------
arsenm wrote:

Can this miss adding other functions in the path to callee that we need to still add to the expression?

e.g. 

```
f1->f2->f3->f2
```

>From f1, we see that callee f2 appears in a recursive expression, so it is skipped. but that will mean we do not add f3 to the expression, even though it is needed 








https://github.com/llvm/llvm-project/pull/112251