param_return: connect wrappers more intelligently

Sometimes wrapper detection might be only partialy right.
In such cases we would like to find wheter arguments
are already handled by analysis and not change them.

For example this will enable test: derived-diamond.arm.clang.O1.elf
in `features/cpp/derived-diamond/symbols`.

src/bin2llvmir/optimizations/param_return/param_return.cpp

@@ -1059,8 +1059,32 @@ void ParamReturn::connectWrappers(const DataFlowEntry& de)
 	unsigned i = 0;
 	for (auto& a : fnc->args())
 	{
-		auto* conv = IrModifier::convertValueToType(&a, wrappedCall->getArgOperand(i)->getType(), wrappedCall);
-		wrappedCall->setArgOperand(i++, conv);
+		auto iarg = wrappedCall->getArgOperand(i);
+		bool shouldSkip = false;
+		if (auto* load = dyn_cast<LoadInst>(llvm_utils::skipCasts(iarg))) {
+			auto oldarg = load->getPointerOperand();
+
+			std::vector<StoreInst*> users;
+			for (const auto& U : oldarg->users())
+			{
+				if (auto* store = dyn_cast<StoreInst>(U)) {
+					if (store->getFunction() == fnc)
+						users.push_back(store);
+				}
+			}
+			for (auto store: users) {
+				if (llvm_utils::skipCasts(store->getValueOperand()) == &a)
+					continue;
+
+				shouldSkip = true;
+			}
+		}
+
+		if (!shouldSkip) {
+			auto* conv = IrModifier::convertValueToType(&a, wrappedCall->getArgOperand(i)->getType(), wrappedCall);
+			wrappedCall->setArgOperand(i, conv);
+		}
+		i++;
 	}
 
 	//