[CIR] Upstream trivial constructor const handling

[andykaylor]

This adds handling for records with trivial constructors in CIR's ConstExprEmitter.

[llvmbot]

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-clangir

Author: Andy Kaylor (andykaylor)

Changes

This adds handling for records with trivial constructors in CIR's ConstExprEmitter.

---

Full diff: https://github.com/llvm/llvm-project/pull/164849.diff

3 Files Affected:

• (modified) clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp (+1-2)
• (modified) clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp (+21-3)
• (added) clang/test/CIR/CodeGen/trivial-ctor-const-init.cpp (+44)

diff --git a/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp b/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp
index d6d226b3f75db..8fe0d9b4a69ef 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp
@@ -362,8 +362,7 @@ class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
     cgf.cgm.errorNYI(e->getSourceRange(), "AggExprEmitter: VisitCXXTypeidExpr");
   }
   void VisitMaterializeTemporaryExpr(MaterializeTemporaryExpr *e) {
-    cgf.cgm.errorNYI(e->getSourceRange(),
-                     "AggExprEmitter: VisitMaterializeTemporaryExpr");
+    Visit(e->getSubExpr());
   }
   void VisitOpaqueValueExpr(OpaqueValueExpr *e) {
     cgf.cgm.errorNYI(e->getSourceRange(),
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp b/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
index 800262aac8fa4..c1f4fe5973d17 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
@@ -1069,9 +1069,27 @@ class ConstExprEmitter
 
   mlir::Attribute VisitCXXConstructExpr(CXXConstructExpr *e, QualType ty) {
     if (!e->getConstructor()->isTrivial())
-      return nullptr;
-    cgm.errorNYI(e->getBeginLoc(), "trivial constructor const handling");
-    return {};
+      return {};
+
+    // Only default and copy/move constructors can be trivial.
+    if (e->getNumArgs()) {
+      assert(e->getNumArgs() == 1 && "trivial ctor with > 1 argument");
+      assert(e->getConstructor()->isCopyOrMoveConstructor() &&
+             "trivial ctor has argument but isn't a copy/move ctor");
+
+      Expr *arg = e->getArg(0);
+      assert(cgm.getASTContext().hasSameUnqualifiedType(ty, arg->getType()) &&
+             "argument to copy ctor is of wrong type");
+
+      // Look through the temporary; it's just converting the value to an lvalue
+      // to pass it to the constructor.
+      if (auto const *mte = dyn_cast<MaterializeTemporaryExpr>(arg))
+        return Visit(mte->getSubExpr(), ty);
+      // Don't try to support arbitrary lvalue-to-rvalue conversions for now.
+      return {};
+    }
+
+    return cgm.getBuilder().getZeroInitAttr(cgm.convertType(ty));
   }
 
   mlir::Attribute VisitStringLiteral(StringLiteral *e, QualType t) {
diff --git a/clang/test/CIR/CodeGen/trivial-ctor-const-init.cpp b/clang/test/CIR/CodeGen/trivial-ctor-const-init.cpp
new file mode 100644
index 0000000000000..7429549100362
--- /dev/null
+++ b/clang/test/CIR/CodeGen/trivial-ctor-const-init.cpp
@@ -0,0 +1,44 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -std=c++11 -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s --check-prefix=CIR
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -std=c++11 -fclangir -emit-llvm %s -o %t-cir.ll
+// RUN: FileCheck --input-file=%t-cir.ll %s --check-prefix=LLVM
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -std=c++11 -emit-llvm %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s --check-prefix=OGCG
+
+struct StructWithDefaultCtor {
+  int n;
+};
+
+StructWithDefaultCtor defCtor = StructWithDefaultCtor();
+
+// CIR: cir.global {{.*}} @defCtor = #cir.zero : !rec_StructWithDefaultCtor
+// LLVM: @defCtor = global %struct.StructWithDefaultCtor zeroinitializer
+// OGCG: @defCtor = global %struct.StructWithDefaultCtor zeroinitializer
+
+struct StructWithCtorArg {
+  double value;
+  StructWithCtorArg(const double& x) : value(x) {}
+};
+
+StructWithCtorArg withArg = 0.0;
+
+// CIR: cir.global {{.*}} @withArg = #cir.zero : !rec_StructWithCtorArg
+// LLVM: @withArg = global %struct.StructWithCtorArg zeroinitializer
+// OGCG: @withArg = global %struct.StructWithCtorArg zeroinitializer
+
+// CIR: cir.func {{.*}} @__cxx_global_var_init()
+// CIR:   %[[TMP0:.*]] = cir.alloca !cir.double, !cir.ptr<!cir.double>, ["ref.tmp0"]
+// CIR:   %[[WITH_ARG:.*]] = cir.get_global @withArg : !cir.ptr<!rec_StructWithCtorArg>
+// CIR:   %[[ZERO:.*]] = cir.const #cir.fp<0.000000e+00> : !cir.double
+// CIR:   cir.store{{.*}} %[[ZERO]], %[[TMP0]] : !cir.double, !cir.ptr<!cir.double>
+// CIR:   cir.call @_ZN17StructWithCtorArgC1ERKd(%[[WITH_ARG]], %[[TMP0]]) : (!cir.ptr<!rec_StructWithCtorArg>, !cir.ptr<!cir.double>) -> ()
+
+// LLVM: define {{.*}} void @__cxx_global_var_init()
+// LLVM:   %[[TMP0:.*]] = alloca double
+// LLVM:   store double 0.000000e+00, ptr %[[TMP0]]
+// LLVM:   call void @_ZN17StructWithCtorArgC1ERKd(ptr @withArg, ptr %[[TMP0]])
+
+// OGCG: define {{.*}} void @__cxx_global_var_init()
+// OGCG:   %[[TMP0:.*]] = alloca double
+// OGCG:   store double 0.000000e+00, ptr %[[TMP0]]
+// OGCG:   call void @_ZN17StructWithCtorArgC1ERKd(ptr {{.*}} @withArg, ptr {{.*}} %[[TMP0]])

[erichkeane]

Should this be a not-yet-implemented or something?

[andykaylor]

This comment is carried over from classic codegen, which does the same thing here.

[erichkeane]

Oh weird... Can you replace the dyn_cast with a cast in classic codegen? Does it hit anything? I'm actually pretty concerned by this...

[andykaylor]

If I report and error after the dyn_cast falls through in classic codegen, it fires 14 times in eight tests in the CodeGenCXX tests:

llvm-project/clang/test/CodeGenCXX/microsoft-uuidof.cpp

Line 44 in a909ec6

GUID thing = (side_effect(), __uuidof(Curly));

llvm-project/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

Line 214 in a909ec6

target({ inner, { destroyme1() } });

llvm-project/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

Line 245 in a909ec6

(void) new std::initializer_list<int> {i, 2, 3};

llvm-project/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

Line 250 in a909ec6

(void) new std::initializer_list<destroyme1> {destroyme1(), destroyme1()};

llvm-project/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

Line 304 in a909ec6

std::initializer_list<S>{ S(), S() };

llvm-project/clang/test/CodeGenCXX/cxx0x-initializer-stdinitializerlist.cpp

Line 321 in a909ec6

auto x = std::initializer_list<S>{ S(), S() };

llvm-project/clang/test/CodeGenCXX/atomicinit.cpp

Line 117 in a909ec6

Y y2 = { x, z };

llvm-project/clang/test/CodeGenCXX/reference-init.cpp

Line 38 in a909ec6

SelfReference self_reference_2 = {self_reference_1};

llvm-project/clang/test/CodeGenCXX/aix-static-init-temp-spec-and-inline-var.cpp

Line 44 in a909ec6

A<int> A<int>::instance = bar();

llvm-project/clang/test/CodeGenCXX/cxx1z-decomposition.cpp

Line 34 in a909ec6

auto [a1, a2] = make<A>();

llvm-project/clang/test/CodeGenCXX/cxx1z-decomposition.cpp

Line 122 in a909ec6

static auto [x1, x2] = make<A>();

llvm-project/clang/test/CodeGenCXX/cxx1z-decomposition.cpp

Line 138 in a909ec6

static auto [x1, x2] = make<B>();

llvm-project/clang/test/CodeGenCXX/static-init-3.cpp

Line 27 in a909ec6

X2< B > bg = X1< X2< B > >::get();

llvm-project/clang/test/CodeGenCXX/copy-constructor-synthesis-2.cpp

Line 11 in a909ec6

PR23373 pr23373_b(pr23373_a);

I haven't looked at these in detail yet, but I know that we can get here if the constructor argument is not a constant, and if we do then returning nullptr is the correct behavior.

The only case that would be a problem is if the argument is not a MaterializeTemporaryExpr but is a constant, and even in that case if we return null here, something else is going to handle it, though possibly not correctly.

I also tried inserting a call to ConstantExprEmitter::Visit(Arg, Ty) in the case where the dyn_cast failed, and that doesn't change anything.

[bcardosolopes]

One possible alternative is to add a missing feature as a way to track what needs to be investigated into both CodeGen paths.

[bcardosolopes]

LGTM once Erich is happy

[bcardosolopes]

One possible alternative is to add a missing feature as a way to track what needs to be investigated into both CodeGen paths.

[erichkeane]

The TODO + missing-features is good enough here, the fact that upstream survived with this being this way for this long means it is probably not pressing. But please create an issue for someone to investigate, obviously minor priority.

[andykaylor]

But please create an issue for someone to investigate, obviously minor priority.

#165017