[Mlir-commits] [flang] [mlir] [mlir][OpenMP] Model memory effects of omp.section and omp.sections (PR #230167)

[llvmlistbot at llvm.org]

================
@@ -3536,6 +3536,29 @@ findEnclosingParallelOp(fir::FirOpBuilder &builder) {
   return {};
 }
 
+static bool isNonDerived(const semantics::Symbol &sym) {
+  const auto *type = sym.GetUltimate().GetType();
+  return type && type->AsIntrinsic();
+}
+
+static bool isNonDerivedItem(const semantics::Symbol *sym) {
+  const auto &ultimate = sym->GetUltimate();
+  // May be a common block
+  if (const auto *common =
+          ultimate.detailsIf<semantics::CommonBlockDetails>()) {
+    return llvm::all_of(common->objects(), [](const auto &member) {
+      return isNonDerived(*member);
+    });
+  }
+  return isNonDerived(ultimate);
+}
+
+static bool isDerivedVariable(mlir::Value variable) {
+  mlir::Type type = variable.getType();
+  return !fir::isa_derived(fir::getFortranElementType(type)) &&
+         !fir::isa_volatile_type(type);
----------------
JRP-arm wrote:

My thinking was that we should be more conservative with volatile types but that's not really the case.

https://github.com/llvm/llvm-project/pull/230167