[CIR] Handle empty unions in record lowering

[Men-cotton]

Handles empty unions (cir.record<union {}>) by generating empty LLVM structs. This ensures successful lowering without triggering crashes in getLargestMember due to missing members.

Added regression test in clang/test/CIR/Lowering/union.cir.

[llvmbot]

@llvm/pr-subscribers-clangir

@llvm/pr-subscribers-clang

Author: Akimasa Watanuki (Men-cotton)

Changes

Handles empty unions (cir.record<union {}>) by generating empty LLVM structs. This ensures successful lowering without triggering crashes in getLargestMember due to missing members.

Added regression test in clang/test/CIR/Lowering/union.cir.

---

Full diff: https://github.com/llvm/llvm-project/pull/172666.diff

2 Files Affected:

• (modified) clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp (+2)
• (added) clang/test/CIR/Lowering/union.cir (+10)

diff --git a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
index 7d854997848aa..06cce00a9aca4 100644
--- a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
+++ b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
@@ -2969,6 +2969,8 @@ static void prepareTypeConverter(mlir::LLVMTypeConverter &converter,
       break;
     // Unions are lowered as only the largest member.
     case cir::RecordType::Union:
+      if (type.getMembers().empty())
+        break;
       if (auto largestMember = type.getLargestMember(dataLayout))
         llvmMembers.push_back(
             convertTypeForMemory(converter, dataLayout, largestMember));
diff --git a/clang/test/CIR/Lowering/union.cir b/clang/test/CIR/Lowering/union.cir
new file mode 100644
index 0000000000000..c7bb8efccadcd
--- /dev/null
+++ b/clang/test/CIR/Lowering/union.cir
@@ -0,0 +1,10 @@
+// RUN: cir-opt %s --cir-to-llvm | FileCheck %s
+
+!u_empty = !cir.record<union {}>
+
+module {
+  cir.func @empty_union(%u: !u_empty) {
+    // CHECK-LABEL:   llvm.func @empty_union
+    cir.return
+  }
+}

[andykaylor]

Can you add a check for the LLVM IR type here? I think we're probably generating the wrong thing. In the incubator, we generate this:

%union.U = type {}

Whereas classic codegen generates this:

%union.U = type { [4 x i8] }

[andykaylor]

Let me amend what I said above. I didn't realize the classic codegen I was referring to was using the Windows C ABI. Classic codegen does generate %union.U = type {} for C and %union.U = type { i8 } for C++ (which the incubator also does).

So, I guess my question is whether we identify these padded forms as "empty".