[CIR] Lower calls to trivial copy constructor to cir::CopyOp (#168281)

This PR is a follow up to #167975 and replaces calls to trivial copy constructors with `cir::CopyOp`. --------- Co-authored-by: Andy Kaylor <akaylor@nvidia.com> Co-authored-by: Henrich Lauko <henrich.lau@gmail.com>
2026-01-17 14:48:27 +08:00 · 2025-12-03 00:22:46 +01:00
parent dbb702fbcb
commit e5f1d025aa
8 changed files with 73 additions and 45 deletions
--- a/clang/lib/CIR/Dialect/Transforms/LoweringPrepare.cpp
+++ b/clang/lib/CIR/Dialect/Transforms/LoweringPrepare.cpp
@@ -74,6 +74,7 @@ struct LoweringPreparePass
  void lowerDynamicCastOp(cir::DynamicCastOp op);
  void lowerArrayDtor(cir::ArrayDtor op);
  void lowerArrayCtor(cir::ArrayCtor op);
+  void lowerTrivialCopyCall(cir::CallOp op);

  /// Build the function that initializes the specified global
  cir::FuncOp buildCXXGlobalVarDeclInitFunc(cir::GlobalOp op);
@@ -1086,6 +1087,25 @@ void LoweringPreparePass::lowerArrayCtor(cir::ArrayCtor op) {
                             true);
 }

+void LoweringPreparePass::lowerTrivialCopyCall(cir::CallOp op) {
+  cir::FuncOp funcOp = getCalledFunction(op);
+  if (!funcOp)
+    return;
+
+  std::optional<cir::CtorKind> ctorKind = funcOp.getCxxConstructorKind();
+  if (ctorKind && *ctorKind == cir::CtorKind::Copy &&
+      funcOp.isCxxTrivialMemberFunction()) {
+    // Replace the trivial copy constructor call with a `CopyOp`
+    CIRBaseBuilderTy builder(getContext());
+    mlir::ValueRange operands = op.getOperands();
+    mlir::Value dest = operands[0];
+    mlir::Value src = operands[1];
+    builder.setInsertionPoint(op);
+    builder.createCopy(dest, src);
+    op.erase();
+  }
+}
+
 void LoweringPreparePass::runOnOp(mlir::Operation *op) {
  if (auto arrayCtor = dyn_cast<cir::ArrayCtor>(op)) {
    lowerArrayCtor(arrayCtor);
@@ -1103,6 +1123,8 @@ void LoweringPreparePass::runOnOp(mlir::Operation *op) {
    lowerDynamicCastOp(dynamicCast);
  } else if (auto unary = mlir::dyn_cast<cir::UnaryOp>(op)) {
    lowerUnaryOp(unary);
+  } else if (auto callOp = dyn_cast<cir::CallOp>(op)) {
+    lowerTrivialCopyCall(callOp);
  } else if (auto fnOp = dyn_cast<cir::FuncOp>(op)) {
    if (auto globalCtor = fnOp.getGlobalCtorPriority())
      globalCtorList.emplace_back(fnOp.getName(), globalCtor.value());
@@ -1121,7 +1143,7 @@ void LoweringPreparePass::runOnOperation() {
  op->walk([&](mlir::Operation *op) {
    if (mlir::isa<cir::ArrayCtor, cir::ArrayDtor, cir::CastOp,
                  cir::ComplexMulOp, cir::ComplexDivOp, cir::DynamicCastOp,
-                  cir::FuncOp, cir::GlobalOp, cir::UnaryOp>(op))
+                  cir::FuncOp, cir::CallOp, cir::GlobalOp, cir::UnaryOp>(op))
      opsToTransform.push_back(op);
  });

--- a/clang/test/CIR/CodeGen/cxx-special-member-attr.cpp
+++ b/clang/test/CIR/CodeGen/cxx-special-member-attr.cpp
@@ -3,25 +3,15 @@

 struct Flub {
  int a = 123;
-  // CIR: @_ZN4FlubC1ERKS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Flub, copy, trivial true>>
-  // CIR: @_ZN4FlubC2EOS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Flub, move, trivial true>
-  // CIR: @_ZN4FlubaSERKS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) -> !cir.ptr<!rec_Flub> special_member<#cir.cxx_assign<!rec_Flub, copy, trivial true>>
-  // CIR: @_ZN4FlubaSEOS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) -> !cir.ptr<!rec_Flub> special_member<#cir.cxx_assign<!rec_Flub, move, trivial true>>
 };

 struct Foo {
  int a;

-  // CIR: @_ZN3FooC2Ev(%arg0: !cir.ptr<!rec_Foo> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Foo, default>>
  Foo() : a(123) {}
-
-  // CIR: @_ZN3FooC2ERKS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Foo, copy>>
  Foo(const Foo &other) : a(other.a) {}
-
-  // CIR: @_ZN3FooC2EOS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Foo, move>>
  Foo(Foo &&other) noexcept : a(other.a) { other.a = 0; }

-  // CIR: @_ZN3FooaSERKS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) -> !cir.ptr<!rec_Foo> special_member<#cir.cxx_assign<!rec_Foo, copy>>
  Foo &operator=(const Foo &other) {
    if (this != &other) {
      a = other.a;
@@ -29,7 +19,6 @@ struct Foo {
    return *this;
  }

-  // CIR: @_ZN3FooaSEOS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) -> !cir.ptr<!rec_Foo> special_member<#cir.cxx_assign<!rec_Foo, move>>
  Foo &operator=(Foo &&other) noexcept {
    if (this != &other) {
      a = other.a;
@@ -38,22 +27,40 @@ struct Foo {
    return *this;
  }

-  // CIR: @_ZN3FooD1Ev(!cir.ptr<!rec_Foo>) special_member<#cir.cxx_dtor<!rec_Foo>>
  ~Foo();
 };

-void trivial() {
+void trivial_func() {
  Flub f1{};
+
  Flub f2 = f1;
+  // Trivial copy constructors/assignments are replaced with cir.copy
+  // CIR: cir.copy {{.*}} : !cir.ptr<!rec_Flub>
+
  Flub f3 = static_cast<Flub&&>(f1);
+  // CIR: @_ZN4FlubC1EOS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Flub, move, trivial true>
+
  f2 = f1;
+  // CIR: @_ZN4FlubaSERKS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) -> !cir.ptr<!rec_Flub> special_member<#cir.cxx_assign<!rec_Flub, copy, trivial true>>
+
  f1 = static_cast<Flub&&>(f3);
+  // CIR: @_ZN4FlubaSEOS_(%arg0: !cir.ptr<!rec_Flub> loc({{.*}}), %arg1: !cir.ptr<!rec_Flub> loc({{.*}})) -> !cir.ptr<!rec_Flub> special_member<#cir.cxx_assign<!rec_Flub, move, trivial true>>
 }

-void non_trivial() {
+void non_trivial_func() {
  Foo f1{};
+  // CIR: @_ZN3FooC2Ev(%arg0: !cir.ptr<!rec_Foo> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Foo, default>>
+
  Foo f2 = f1;
+  // CIR: @_ZN3FooC2ERKS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Foo, copy>>
+
  Foo f3 = static_cast<Foo&&>(f1);
+  // CIR: @_ZN3FooC2EOS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) special_member<#cir.cxx_ctor<!rec_Foo, move>>
+
  f2 = f1;
+  // CIR: @_ZN3FooaSERKS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) -> !cir.ptr<!rec_Foo> special_member<#cir.cxx_assign<!rec_Foo, copy>>
+
  f1 = static_cast<Foo&&>(f3);
+  // CIR: @_ZN3FooaSEOS_(%arg0: !cir.ptr<!rec_Foo> loc({{.*}}), %arg1: !cir.ptr<!rec_Foo> loc({{.*}})) -> !cir.ptr<!rec_Foo> special_member<#cir.cxx_assign<!rec_Foo, move>>
+  // CIR: @_ZN3FooD1Ev(!cir.ptr<!rec_Foo>) special_member<#cir.cxx_dtor<!rec_Foo>>
 }
--- a/clang/test/CIR/CodeGen/lambda.cpp
+++ b/clang/test/CIR/CodeGen/lambda.cpp
@@ -367,7 +367,7 @@ struct A {
 // CIR:   cir.scope {
 // CIR:     %[[LAM_ADDR:.*]] = cir.alloca ![[REC_LAM_A]], !cir.ptr<![[REC_LAM_A]]>, ["ref.tmp0"]
 // CIR:     %[[STRUCT_A:.*]] = cir.get_member %[[LAM_ADDR]][0] {name = "this"} : !cir.ptr<![[REC_LAM_A]]> -> !cir.ptr<!rec_A>
-// CIR:     cir.call @_ZN1AC1ERKS_(%[[STRUCT_A]], %[[THIS]]){{.*}} : (!cir.ptr<!rec_A>, !cir.ptr<!rec_A>){{.*}} -> ()
+// CIR:     cir.copy %[[THIS]] to %[[STRUCT_A]] : !cir.ptr<!rec_A>
 // CIR:     %[[LAM_RET:.*]] = cir.call @_ZZN1A3fooEvENKUlvE_clEv(%[[LAM_ADDR]])
 // CIR:     cir.store{{.*}} %[[LAM_RET]], %[[RETVAL]]
 // CIR:   }
@@ -383,7 +383,7 @@ struct A {
 // LLVM:   br label %[[SCOPE_BB:.*]]
 // LLVM: [[SCOPE_BB]]:
 // LLVM:   %[[STRUCT_A:.*]] = getelementptr %[[REC_LAM_A]], ptr %[[LAM_ALLOCA]], i32 0, i32 0
-// LLVM:   call void @_ZN1AC1ERKS_(ptr %[[STRUCT_A]], ptr %[[THIS]])
+// LLVM:   call void @llvm.memcpy.p0.p0.i32(ptr %[[STRUCT_A]], ptr %[[THIS]], i32 4, i1 false)
 // LLVM:   %[[LAM_RET:.*]] = call i32 @_ZZN1A3fooEvENKUlvE_clEv(ptr %[[LAM_ALLOCA]])
 // LLVM:   store i32 %[[LAM_RET]], ptr %[[RETVAL]]
 // LLVM:   br label %[[RET_BB:.*]]
--- a/clang/test/CIR/CodeGen/struct.cpp
+++ b/clang/test/CIR/CodeGen/struct.cpp
@@ -109,13 +109,13 @@ void paren_expr() {
 // CIR:   %[[B_ADDR:.*]] = cir.alloca !rec_Point, !cir.ptr<!rec_Point>, ["b", init]
 // CIR:   %[[CONST:.*]] = cir.const #cir.zero : !rec_Point
 // CIR:   cir.store{{.*}} %[[CONST]], %[[A_ADDR]] : !rec_Point, !cir.ptr<!rec_Point>
-// CIR:   cir.call @_ZZ10paren_exprvEN5PointC1ERKS_(%[[B_ADDR]], %[[A_ADDR]]) nothrow : (!cir.ptr<!rec_Point>, !cir.ptr<!rec_Point>) -> ()
+// CIR:   cir.copy %[[A_ADDR]] to %[[B_ADDR]] : !cir.ptr<!rec_Point>

 // LLVM: define{{.*}} void @_Z10paren_exprv()
 // LLVM:   %[[A_ADDR:.*]] = alloca %struct.Point, i64 1, align 4
 // LLVM:   %[[B_ADDR:.*]] = alloca %struct.Point, i64 1, align 4
 // LLVM:   store %struct.Point zeroinitializer, ptr %[[A_ADDR]], align 4
-// LLVM:   call void @_ZZ10paren_exprvEN5PointC1ERKS_(ptr %[[B_ADDR]], ptr %[[A_ADDR]])
+// LLVM:   call void @llvm.memcpy.p0.p0.i32(ptr %[[B_ADDR]], ptr %[[A_ADDR]], i32 8, i1 false)

 // OGCG: define{{.*}} void @_Z10paren_exprv()
 // OGCG:   %[[A_ADDR:.*]] = alloca %struct.Point, align 4
@@ -133,14 +133,13 @@ void choose_expr() {
 // CIR:   %[[A_ADDR:.*]] = cir.alloca !rec_CompleteS, !cir.ptr<!rec_CompleteS>, ["a"]
 // CIR:   %[[B_ADDR:.*]] = cir.alloca !rec_CompleteS, !cir.ptr<!rec_CompleteS>, ["b"]
 // CIR:   %[[C_ADDR:.*]] = cir.alloca !rec_CompleteS, !cir.ptr<!rec_CompleteS>, ["c", init]
-// TODO(cir): Call to default copy constructor should be replaced by `cir.copy` op
-// CIR:   cir.call @_ZN9CompleteSC1ERKS_(%[[C_ADDR]], %[[A_ADDR]]) nothrow : (!cir.ptr<!rec_CompleteS>, !cir.ptr<!rec_CompleteS>) -> ()
+// CIR:   cir.copy %[[A_ADDR]] to %[[C_ADDR]] : !cir.ptr<!rec_CompleteS>

 // LLVM: define{{.*}} void @_Z11choose_exprv()
 // LLVM:   %[[A_ADDR:.*]] = alloca %struct.CompleteS, i64 1, align 4
 // LLVM:   %[[B_ADDR:.*]] = alloca %struct.CompleteS, i64 1, align 4
 // LLVM:   %[[C_ADDR:.*]] = alloca %struct.CompleteS, i64 1, align 4
-// LLVM:   call void @_ZN9CompleteSC1ERKS_(ptr %[[C_ADDR]], ptr %[[A_ADDR]])
+// LLVM:   call void @llvm.memcpy.p0.p0.i32(ptr %[[C_ADDR]], ptr %[[A_ADDR]], i32 8, i1 false)

 // OGCG: define{{.*}} void @_Z11choose_exprv()
 // OGCG:   %[[A_ADDR:.*]] = alloca %struct.CompleteS, align 4
@@ -160,15 +159,14 @@ void generic_selection() {
 // CIR:   %[[B_ADDR:.*]] = cir.alloca !rec_CompleteS, !cir.ptr<!rec_CompleteS>, ["b"]
 // CIR:   %[[C_ADDR:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["c"]
 // CIR:   %[[D_ADDR:.*]] = cir.alloca !rec_CompleteS, !cir.ptr<!rec_CompleteS>, ["d", init]
-// TODO(cir): Call to default copy constructor should be replaced by `cir.copy` op
-// CIR:   cir.call @_ZN9CompleteSC1ERKS_(%[[D_ADDR]], %[[A_ADDR]]) nothrow : (!cir.ptr<!rec_CompleteS>, !cir.ptr<!rec_CompleteS>) -> ()
+// CIR:   cir.copy %[[A_ADDR]] to %[[D_ADDR]] : !cir.ptr<!rec_CompleteS>

 // LLVM: define{{.*}} void @_Z17generic_selectionv()
 // LLVM:   %1 = alloca %struct.CompleteS, i64 1, align 4
 // LLVM:   %2 = alloca %struct.CompleteS, i64 1, align 4
 // LLVM:   %3 = alloca i32, i64 1, align 4
 // LLVM:   %4 = alloca %struct.CompleteS, i64 1, align 4
-// LLVM:   call void @_ZN9CompleteSC1ERKS_(ptr %4, ptr %1)
+// LLVM:   call void @llvm.memcpy.p0.p0.i32(ptr %4, ptr %1, i32 8, i1 false)

 // OGCG: define{{.*}} void @_Z17generic_selectionv()
 // OGCG:   %[[A_ADDR:.*]] = alloca %struct.CompleteS, align 4
@@ -188,7 +186,7 @@ void designated_init_update_expr() {
 // CIR: %[[A_ADDR:.*]] = cir.alloca !rec_CompleteS, !cir.ptr<!rec_CompleteS>, ["a"]
 // CIR: %[[B_ADDR:.*]] = cir.alloca !rec_Container, !cir.ptr<!rec_Container>, ["b", init]
 // CIR: %[[C_ADDR:.*]] = cir.get_member %[[B_ADDR]][0] {name = "c"} : !cir.ptr<!rec_Container> -> !cir.ptr<!rec_CompleteS>
-// CIR: cir.call @_ZN9CompleteSC1ERKS_(%2, %[[A_ADDR]]) nothrow : (!cir.ptr<!rec_CompleteS>, !cir.ptr<!rec_CompleteS>) -> ()
+// CIR: cir.copy %[[A_ADDR]] to %[[C_ADDR]] : !cir.ptr<!rec_CompleteS>
 // CIR: %[[ELEM_0_PTR:.*]] = cir.get_member %[[C_ADDR]][0] {name = "a"} : !cir.ptr<!rec_CompleteS> -> !cir.ptr<!s32i>
 // CIR: %[[CONST_1:.*]] = cir.const #cir.int<1> : !s32i
 // CIR: cir.store{{.*}} %[[CONST_1]], %[[ELEM_0_PTR]] : !s32i, !cir.ptr<!s32i>
@@ -197,7 +195,7 @@ void designated_init_update_expr() {
 // LLVM: %[[A_ADDR:.*]] = alloca %struct.CompleteS, i64 1, align 4
 // LLVM: %[[B_ADDR:.*]] = alloca %struct.Container, i64 1, align 4
 // LLVM: %[[C_ADDR:.*]] = getelementptr %struct.Container, ptr %[[B_ADDR]], i32 0, i32 0
-// LLVM: call void @_ZN9CompleteSC1ERKS_(ptr %[[C_ADDR]], ptr %[[A_ADDR]])
+// LLVM: call void @llvm.memcpy.p0.p0.i32(ptr %[[C_ADDR]], ptr %[[A_ADDR]], i32 8, i1 false)
 // LLVM: %[[ELEM_0_PTR:.*]] = getelementptr %struct.CompleteS, ptr %[[C_ADDR]], i32 0, i32 0
 // LLVM: store i32 1, ptr %[[ELEM_0_PTR]], align 4
 // LLVM: %[[ELEM_1_PTR:.*]] = getelementptr %struct.CompleteS, ptr %[[C_ADDR]], i32 0, i32 1
--- a/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
@@ -43,7 +43,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_NoCopyConstruct> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_NoCopyConstruct> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN15NoCopyConstructC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_NoCopyConstruct>, !cir.ptr<!rec_NoCopyConstruct>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
 //
@@ -63,7 +63,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_NonDefaultCtor> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_NonDefaultCtor> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
 //
@@ -73,7 +73,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
 // CHECK-NEXT: ^bb0(%[[ORIG:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}, %[[ARG:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}):
@@ -176,7 +176,7 @@ struct HasDtor {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[DECAY_FROM]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NoCopyConstruct>, !u64i) -> !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>> -> !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[DECAY_TO]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NoCopyConstruct>, !u64i) -> !cir.ptr<!rec_NoCopyConstruct>
-// CHECK-NEXT: cir.call @_ZN15NoCopyConstructC1ERKS_(%[[STRIDE_TO]], %[[STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_NoCopyConstruct>, !cir.ptr<!rec_NoCopyConstruct>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR_LOAD]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
@@ -246,7 +246,7 @@ struct HasDtor {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[DECAY_FROM]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NonDefaultCtor>, !u64i) -> !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>> -> !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[DECAY_TO]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NonDefaultCtor>, !u64i) -> !cir.ptr<!rec_NonDefaultCtor>
-// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[STRIDE_TO]], %[[STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR_LOAD]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
@@ -281,7 +281,7 @@ struct HasDtor {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[DECAY_FROM]], %[[ITR_LOAD]] : (!cir.ptr<!rec_HasDtor>, !u64i) -> !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>> -> !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[DECAY_TO]], %[[ITR_LOAD]] : (!cir.ptr<!rec_HasDtor>, !u64i) -> !cir.ptr<!rec_HasDtor>
-// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[STRIDE_TO]], %[[STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR_LOAD]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
--- a/clang/test/CIR/CodeGenOpenACC/compute-firstprivate-clause-templates.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/compute-firstprivate-clause-templates.cpp
@@ -30,7 +30,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_NonDefaultCtor> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_NonDefaultCtor> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
 //
@@ -40,7 +40,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
 // CHECK-NEXT: ^bb0(%[[ORIG:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}, %[[ARG:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}):
--- a/clang/test/CIR/CodeGenOpenACC/compute-firstprivate-clause.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/compute-firstprivate-clause.cpp
@@ -1,4 +1,5 @@
-// RUN: %clang_cc1 -fopenacc -triple x86_64-linux-gnu -Wno-openacc-self-if-potential-conflict -emit-cir -fclangir -triple x86_64-linux-pc %s -o - | FileCheck %s
+// RUN: %clang_cc1 -fopenacc -triple x86_64-linux-gnu -Wno-openacc-self-if-potential-conflict -emit-cir -fclangir -triple x86_64-linux-pc %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s

 struct NoCopyConstruct {};

@@ -43,7 +44,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_NoCopyConstruct> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_NoCopyConstruct> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN15NoCopyConstructC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_NoCopyConstruct>, !cir.ptr<!rec_NoCopyConstruct>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
 //
@@ -63,7 +64,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_NonDefaultCtor> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_NonDefaultCtor> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
 //
@@ -73,7 +74,7 @@ struct HasDtor {
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } copy {
 // CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}):
-// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[ARG_TO]], %[[ARG_FROM]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+// CHECK-NEXT: cir.copy %[[ARG_FROM]] to %[[ARG_TO]] : !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
 // CHECK-NEXT: ^bb0(%[[ORIG:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}, %[[ARG:.*]]: !cir.ptr<!rec_HasDtor> {{.*}}):
@@ -176,7 +177,7 @@ struct HasDtor {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[DECAY_FROM]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NoCopyConstruct>, !u64i) -> !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>> -> !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[DECAY_TO]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NoCopyConstruct>, !u64i) -> !cir.ptr<!rec_NoCopyConstruct>
-// CHECK-NEXT: cir.call @_ZN15NoCopyConstructC1ERKS_(%[[STRIDE_TO]], %[[STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_NoCopyConstruct>, !cir.ptr<!rec_NoCopyConstruct>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_NoCopyConstruct>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR_LOAD]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
@@ -246,7 +247,7 @@ struct HasDtor {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[DECAY_FROM]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NonDefaultCtor>, !u64i) -> !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>> -> !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[DECAY_TO]], %[[ITR_LOAD]] : (!cir.ptr<!rec_NonDefaultCtor>, !u64i) -> !cir.ptr<!rec_NonDefaultCtor>
-// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[STRIDE_TO]], %[[STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_NonDefaultCtor>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR_LOAD]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
@@ -281,7 +282,7 @@ struct HasDtor {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[DECAY_FROM]], %[[ITR_LOAD]] : (!cir.ptr<!rec_HasDtor>, !u64i) -> !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>> -> !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[DECAY_TO]], %[[ITR_LOAD]] : (!cir.ptr<!rec_HasDtor>, !u64i) -> !cir.ptr<!rec_HasDtor>
-// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[STRIDE_TO]], %[[STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_HasDtor>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR_LOAD]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
--- a/clang/test/CIR/CodeGenOpenACC/firstprivate-clause-recipes.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/firstprivate-clause-recipes.cpp
@@ -75,7 +75,7 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[BOUND1_STRIDE_FROM:.*]] = cir.ptr_stride %[[BOUND2_STRIDE_DECAY_FROM]], %[[ITR1_LOAD]] : (!cir.ptr<!rec_NoOps>, !u64i) -> !cir.ptr<!rec_NoOps>
 // CHECK-NEXT: %[[BOUND2_STRIDE_DECAY_TO:.*]] = cir.cast array_to_ptrdecay %[[BOUND2_STRIDE_TO]] : !cir.ptr<!cir.array<!rec_NoOps x 5>> -> !cir.ptr<!rec_NoOps>
 // CHECK-NEXT: %[[BOUND1_STRIDE_TO:.*]] = cir.ptr_stride %[[BOUND2_STRIDE_DECAY_TO]], %[[ITR1_LOAD]] : (!cir.ptr<!rec_NoOps>, !u64i) -> !cir.ptr<!rec_NoOps>
-// CHECK-NEXT: cir.call @_ZN5NoOpsC1ERKS_(%[[BOUND1_STRIDE_TO]], %[[BOUND1_STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_NoOps>, !cir.ptr<!rec_NoOps>) -> ()
+// CHECK-NEXT: cir.copy %[[BOUND1_STRIDE_FROM]] to %[[BOUND1_STRIDE_TO]] : !cir.ptr<!rec_NoOps>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR1_LOAD]] = cir.load %[[ITR1]] : !cir.ptr<!u64i>, !u64i
@@ -342,7 +342,7 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[BOUND2_STRIDE_LOAD_FROM]], %[[ITR1_LOAD]] : (!cir.ptr<!rec_NoOps>, !u64i) -> !cir.ptr<!rec_NoOps>
 // CHECK-NEXT: %[[BOUND2_STRIDE_LOAD_TO:.*]] = cir.load %[[BOUND2_STRIDE_TO]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!rec_NoOps>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[BOUND2_STRIDE_LOAD_TO]], %[[ITR1_LOAD]] : (!cir.ptr<!rec_NoOps>, !u64i) -> !cir.ptr<!rec_NoOps>
-// CHECK-NEXT: cir.call @_ZN5NoOpsC1ERKS_(%[[BOUND1_STRIDE_TO]], %[[BOUND1_STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_NoOps>, !cir.ptr<!rec_NoOps>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_NoOps>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR1_LOAD]] = cir.load %[[ITR1]] : !cir.ptr<!u64i>, !u64i
@@ -580,7 +580,7 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[STRIDE_FROM:.*]] = cir.ptr_stride %[[BOUND2_STRIDE_LOAD_FROM]], %[[ITR1_LOAD]] : (!cir.ptr<!rec_CtorDtor>, !u64i) -> !cir.ptr<!rec_CtorDtor>
 // CHECK-NEXT: %[[BOUND2_STRIDE_LOAD_TO:.*]] = cir.load %[[BOUND2_STRIDE_TO]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!rec_CtorDtor>
 // CHECK-NEXT: %[[STRIDE_TO:.*]] = cir.ptr_stride %[[BOUND2_STRIDE_LOAD_TO]], %[[ITR1_LOAD]] : (!cir.ptr<!rec_CtorDtor>, !u64i) -> !cir.ptr<!rec_CtorDtor>
-// CHECK-NEXT: cir.call @_ZN8CtorDtorC1ERKS_(%[[BOUND1_STRIDE_TO]], %[[BOUND1_STRIDE_FROM]]) nothrow : (!cir.ptr<!rec_CtorDtor>, !cir.ptr<!rec_CtorDtor>) -> ()
+// CHECK-NEXT: cir.copy %[[STRIDE_FROM]] to %[[STRIDE_TO]] : !cir.ptr<!rec_CtorDtor>
 // CHECK-NEXT: cir.yield
 // CHECK-NEXT: } step {
 // CHECK-NEXT: %[[ITR1_LOAD]] = cir.load %[[ITR1]] : !cir.ptr<!u64i>, !u64i