[flang] Avoid generating duplicate symbol in comdat

[clementval]

In case where a fir.global might be duplicated in an inner module (gpu.module), the conversion pattern will be applied on the module and the gpu module version of the global and try to generate multiple comdat with the same symbol name. This is what we have in the implementation of CUDA Fortran.

Just check for the presence of the ComdatSelectorOp before creating a new one.

[llvmbot]

@llvm/pr-subscribers-flang-codegen

Author: Valentin Clement (バレンタイン クレメン) (clementval)

Changes

In case where a fir.global might be duplicated in an inner module (gpu.module), the conversion pattern will be applied on the module and the gpu module version of the global and try to generate multiple comdat with the same symbol name. This is what we have in the implementation of CUDA Fortran.

Just check for the presence of the ComdatSelectorOp before creating a new one.

---

Full diff: https://github.com/llvm/llvm-project/pull/114472.diff

2 Files Affected:

• (modified) flang/lib/Optimizer/CodeGen/CodeGen.cpp (+3)
• (added) flang/test/Fir/comdat-present.fir (+14)

diff --git a/flang/lib/Optimizer/CodeGen/CodeGen.cpp b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
index 4c8c56e0f21cef..d038efcb2eb42c 100644
--- a/flang/lib/Optimizer/CodeGen/CodeGen.cpp
+++ b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
@@ -2931,6 +2931,9 @@ struct GlobalOpConversion : public fir::FIROpConversion<fir::GlobalOp> {
       comdatOp =
           rewriter.create<mlir::LLVM::ComdatOp>(module.getLoc(), comdatName);
     }
+    if (auto select = comdatOp.lookupSymbol<mlir::LLVM::ComdatSelectorOp>(
+            global.getSymName()))
+      return;
     mlir::OpBuilder::InsertionGuard guard(rewriter);
     rewriter.setInsertionPointToEnd(&comdatOp.getBody().back());
     auto selectorOp = rewriter.create<mlir::LLVM::ComdatSelectorOp>(
diff --git a/flang/test/Fir/comdat-present.fir b/flang/test/Fir/comdat-present.fir
new file mode 100644
index 00000000000000..96d14e5973f4f7
--- /dev/null
+++ b/flang/test/Fir/comdat-present.fir
@@ -0,0 +1,14 @@
+// RUN: fir-opt %s --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %s
+// RUN: fir-opt %s --fir-to-llvm-ir="target=x86_64-pc-windows-msvc" | FileCheck %s
+
+fir.global linkonce_odr @global_linkonce_odr constant : i32 {
+  %0 = arith.constant 0 : i32
+  fir.has_value %0 : i32
+}
+
+llvm.comdat @__llvm_comdat {
+  llvm.comdat_selector @global_linkonce_odr any
+}
+
+// CHECK-LABEL: llvm.comdat @__llvm_comdat
+// CHECK: llvm.comdat_selector @global_linkonce_odr any

[llvmbot]

@llvm/pr-subscribers-flang-fir-hlfir

Author: Valentin Clement (バレンタイン クレメン) (clementval)

Changes

In case where a fir.global might be duplicated in an inner module (gpu.module), the conversion pattern will be applied on the module and the gpu module version of the global and try to generate multiple comdat with the same symbol name. This is what we have in the implementation of CUDA Fortran.

Just check for the presence of the ComdatSelectorOp before creating a new one.

---

Full diff: https://github.com/llvm/llvm-project/pull/114472.diff

2 Files Affected:

• (modified) flang/lib/Optimizer/CodeGen/CodeGen.cpp (+3)
• (added) flang/test/Fir/comdat-present.fir (+14)

diff --git a/flang/lib/Optimizer/CodeGen/CodeGen.cpp b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
index 4c8c56e0f21cef..d038efcb2eb42c 100644
--- a/flang/lib/Optimizer/CodeGen/CodeGen.cpp
+++ b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
@@ -2931,6 +2931,9 @@ struct GlobalOpConversion : public fir::FIROpConversion<fir::GlobalOp> {
       comdatOp =
           rewriter.create<mlir::LLVM::ComdatOp>(module.getLoc(), comdatName);
     }
+    if (auto select = comdatOp.lookupSymbol<mlir::LLVM::ComdatSelectorOp>(
+            global.getSymName()))
+      return;
     mlir::OpBuilder::InsertionGuard guard(rewriter);
     rewriter.setInsertionPointToEnd(&comdatOp.getBody().back());
     auto selectorOp = rewriter.create<mlir::LLVM::ComdatSelectorOp>(
diff --git a/flang/test/Fir/comdat-present.fir b/flang/test/Fir/comdat-present.fir
new file mode 100644
index 00000000000000..96d14e5973f4f7
--- /dev/null
+++ b/flang/test/Fir/comdat-present.fir
@@ -0,0 +1,14 @@
+// RUN: fir-opt %s --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %s
+// RUN: fir-opt %s --fir-to-llvm-ir="target=x86_64-pc-windows-msvc" | FileCheck %s
+
+fir.global linkonce_odr @global_linkonce_odr constant : i32 {
+  %0 = arith.constant 0 : i32
+  fir.has_value %0 : i32
+}
+
+llvm.comdat @__llvm_comdat {
+  llvm.comdat_selector @global_linkonce_odr any
+}
+
+// CHECK-LABEL: llvm.comdat @__llvm_comdat
+// CHECK: llvm.comdat_selector @global_linkonce_odr any

[Renaud-K]

Thank you.