[OPENMP] Fixed codegen for copying/initialization of array variables/parameters.

This modification generates proper copyin/initialization sequences for array variables/parameters. Before they were considered as pointers, not arrays.

llvm-svn: 237691

clang/lib/CodeGen/CGOpenMPRuntime.cpp

@@ -1287,7 +1287,12 @@ static llvm::Value *emitCopyprivateCopyFunction(
             CGF.Builder.CreateStructGEP(nullptr, RHS, I),
             CGM.PointerAlignInBytes),
         CGF.ConvertTypeForMem(C.getPointerType(SrcExprs[I]->getType())));
-    CGF.EmitOMPCopy(CGF, CopyprivateVars[I]->getType(), DestAddr, SrcAddr,
+    auto *VD = cast<DeclRefExpr>(CopyprivateVars[I])->getDecl();
+    QualType Type = VD->getType();
+    if (auto *PVD = dyn_cast<ParmVarDecl>(VD)) {
+      Type = PVD->getOriginalType();
+    }
+    CGF.EmitOMPCopy(CGF, Type, DestAddr, SrcAddr,
                     cast<VarDecl>(cast<DeclRefExpr>(DestExprs[I])->getDecl()),
                     cast<VarDecl>(cast<DeclRefExpr>(SrcExprs[I])->getDecl()),
                     AssignmentOps[I]);
@@ -1654,8 +1659,12 @@ createPrivatesRecordDecl(CodeGenModule &CGM,
     auto *RD = C.buildImplicitRecord(".kmp_privates.t");
     RD->startDefinition();
     for (auto &&Pair : Privates) {
-      addFieldToRecordDecl(
+      auto Type = Pair.second.Original->getType();
-          C, RD, Pair.second.Original->getType().getNonReferenceType());
+      if (auto *PVD = dyn_cast<ParmVarDecl>(Pair.second.Original)) {
+        Type = PVD->getOriginalType();
+      }
+      Type = Type.getNonReferenceType();
+      addFieldToRecordDecl(C, RD, Type);
     }
     RD->completeDefinition();
     return RD;
@@ -1939,22 +1948,24 @@ void CGOpenMPRuntime::emitTaskCall(
           auto *SharedField = CapturesInfo.lookup(OriginalVD);
           auto SharedRefLValue =
               CGF.EmitLValueForField(SharedsBase, SharedField);
-          if (OriginalVD->getType()->isArrayType()) {
+          QualType Type = OriginalVD->getType();
+          if (auto *PVD = dyn_cast<ParmVarDecl>(OriginalVD)) {
+            Type = PVD->getOriginalType();
+          }
+          if (Type->isArrayType()) {
             // Initialize firstprivate array.
             if (!isa<CXXConstructExpr>(Init) ||
                 CGF.isTrivialInitializer(Init)) {
               // Perform simple memcpy.
               CGF.EmitAggregateAssign(PrivateLValue.getAddress(),
-                                      SharedRefLValue.getAddress(),
+                                      SharedRefLValue.getAddress(), Type);
-                                      OriginalVD->getType());
             } else {
               // Initialize firstprivate array using element-by-element
               // intialization.
               CGF.EmitOMPAggregateAssign(
                   PrivateLValue.getAddress(), SharedRefLValue.getAddress(),
-                  OriginalVD->getType(),
+                  Type, [&CGF, Elem, Init, &CapturesInfo](
-                  [&CGF, Elem, Init, &CapturesInfo](llvm::Value *DestElement,
+                            llvm::Value *DestElement, llvm::Value *SrcElement) {
-                                                    llvm::Value *SrcElement) {
                     // Clean up any temporaries needed by the initialization.
                     CodeGenFunction::OMPPrivateScope InitScope(CGF);
                     InitScope.addPrivate(Elem, [SrcElement]() -> llvm::Value *{
@@ -1996,8 +2007,12 @@ void CGOpenMPRuntime::emitTaskCall(
       auto *SharedField = CapturesInfo.lookup(OriginalVD);
       auto SharedRefLValue =
           CGF.EmitLValueForFieldInitialization(SharedsBase, SharedField);
-      CGF.EmitStoreThroughLValue(RValue::get(PrivateLValue.getAddress()),
+      CGF.EmitStoreThroughLValue(
-                                 SharedRefLValue);
+          RValue::get(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              PrivateLValue.getAddress(), SharedRefLValue.getAddress()
+                                              ->getType()
+                                              ->getPointerElementType())),
+          SharedRefLValue);
       ++FI, ++I;
     }
   }

clang/lib/CodeGen/CGStmtOpenMP.cpp

@@ -132,7 +132,11 @@ bool CodeGenFunction::EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
                 OrigVD) != nullptr,
             (*IRef)->getType(), VK_LValue, (*IRef)->getExprLoc());
         auto *OriginalAddr = EmitLValue(&DRE).getAddress();
-        if (OrigVD->getType()->isArrayType()) {
+        QualType Type = OrigVD->getType();
+        if (auto *PVD = dyn_cast<ParmVarDecl>(OrigVD)) {
+          Type = PVD->getOriginalType();
+        }
+        if (Type->isArrayType()) {
           // Emit VarDecl with copy init for arrays.
           // Get the address of the original variable captured in current
           // captured region.
@@ -142,11 +146,10 @@ bool CodeGenFunction::EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
             if (!isa<CXXConstructExpr>(Init) || isTrivialInitializer(Init)) {
               // Perform simple memcpy.
               EmitAggregateAssign(Emission.getAllocatedAddress(), OriginalAddr,
-                                  (*IRef)->getType());
+                                  Type);
             } else {
               EmitOMPAggregateAssign(
-                  Emission.getAllocatedAddress(), OriginalAddr,
+                  Emission.getAllocatedAddress(), OriginalAddr, Type,
-                  (*IRef)->getType(),
                   [this, VDInit, Init](llvm::Value *DestElement,
                                        llvm::Value *SrcElement) {
                     // Clean up any temporaries needed by the initialization.
@@ -225,6 +228,10 @@ bool CodeGenFunction::EmitOMPCopyinClause(const OMPExecutableDirective &D) {
     auto IDestRef = C->destination_exprs().begin();
     for (auto *AssignOp : C->assignment_ops()) {
       auto *VD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+      QualType Type = VD->getType();
+      if (auto *PVD = dyn_cast<ParmVarDecl>(VD)) {
+        Type = PVD->getOriginalType();
+      }
       if (CopiedVars.insert(VD->getCanonicalDecl()).second) {
         // Get the address of the master variable.
         auto *MasterAddr = VD->isStaticLocal()
@@ -246,8 +253,8 @@ bool CodeGenFunction::EmitOMPCopyinClause(const OMPExecutableDirective &D) {
         }
         auto *SrcVD = cast<VarDecl>(cast<DeclRefExpr>(*ISrcRef)->getDecl());
         auto *DestVD = cast<VarDecl>(cast<DeclRefExpr>(*IDestRef)->getDecl());
-        EmitOMPCopy(*this, (*IRef)->getType(), PrivateAddr, MasterAddr, DestVD,
+        EmitOMPCopy(*this, Type, PrivateAddr, MasterAddr, DestVD, SrcVD,
-                    SrcVD, AssignOp);
+                    AssignOp);
       }
       ++IRef;
       ++ISrcRef;
@@ -328,6 +335,10 @@ void CodeGenFunction::EmitOMPLastprivateClauseFinal(
       auto IDestRef = C->destination_exprs().begin();
       for (auto *AssignOp : C->assignment_ops()) {
         auto *PrivateVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+        QualType Type = PrivateVD->getType();
+        if (auto *PVD = dyn_cast<ParmVarDecl>(PrivateVD)) {
+          Type = PVD->getOriginalType();
+        }
         if (AlreadyEmittedVars.insert(PrivateVD->getCanonicalDecl()).second) {
           auto *SrcVD = cast<VarDecl>(cast<DeclRefExpr>(*ISrcRef)->getDecl());
           auto *DestVD = cast<VarDecl>(cast<DeclRefExpr>(*IDestRef)->getDecl());
@@ -335,8 +346,8 @@ void CodeGenFunction::EmitOMPLastprivateClauseFinal(
           auto *OriginalAddr = GetAddrOfLocalVar(DestVD);
           // Get the address of the private variable.
           auto *PrivateAddr = GetAddrOfLocalVar(PrivateVD);
-          EmitOMPCopy(*this, (*IRef)->getType(), OriginalAddr, PrivateAddr,
+          EmitOMPCopy(*this, Type, OriginalAddr, PrivateAddr, DestVD, SrcVD,
-                      DestVD, SrcVD, AssignOp);
+                      AssignOp);
         }
         ++IRef;
         ++ISrcRef;

clang/test/OpenMP/parallel_copyin_codegen.cpp

@@ -3,7 +3,9 @@
 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s
 // expected-no-diagnostics
+#ifndef ARRAY
 #ifndef HEADER
 #define HEADER
 
@@ -269,4 +271,26 @@ int main() {
 // CHECK: ret void
 
 #endif
+#else
+// ARRAY-LABEL: array_func
+struct St {
+  int a, b;
+  St() : a(0), b(0) {}
+  St &operator=(const St &) { return *this; };
+  ~St() {}
+};
+
+void array_func() {
+  static int a[2];
+  static St s[2];
+// ARRAY: @__kmpc_fork_call(
+// ARRAY: call i8* @__kmpc_threadprivate_cached(
+// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* bitcast ([2 x i32]* @{{.+}} to i8*), i64 8, i32 4, i1 false)
+// ARRAY: call dereferenceable(8) %struct.St* @{{.+}}(%struct.St* %{{.+}}, %struct.St* dereferenceable(8) %{{.+}})
+#pragma omp threadprivate(a, s)
+#pragma omp parallel copyin(a, s)
+  ;
+}
+#endif
+
 

clang/test/OpenMP/parallel_firstprivate_codegen.cpp

@@ -3,7 +3,9 @@
 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s
 // expected-no-diagnostics
+#ifndef ARRAY
 #ifndef HEADER
 #define HEADER
 
@@ -237,5 +239,24 @@ int main() {
 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]* [[VAR_PRIV]])
 // CHECK-DAG: call {{.*}} [[S_INT_TY_DESTR]]([[S_INT_TY]]*
 // CHECK: ret void
+
+#endif
+#else
+// ARRAY-LABEL: array_func
+struct St {
+  int a, b;
+  St() : a(0), b(0) {}
+  St(const St &) { }
+  ~St() {}
+};
+
+void array_func(int a[3], St s[2]) {
+// ARRAY: @__kmpc_fork_call(
+// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* %{{.+}}, i64 12, i32 4, i1 false)
+// ARRAY: call void @_ZN2StC1ERKS_(%struct.St* %{{.+}}, %struct.St* dereferenceable(8) %{{.+}}
+#pragma omp parallel firstprivate(a, s)
+  ;
+}
 #endif
 
+

clang/test/OpenMP/single_codegen.cpp

@@ -2,8 +2,10 @@
 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp=libiomp5 -fexceptions -fcxx-exceptions -gline-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s
 // expected-no-diagnostics
 
+#ifndef ARRAY
 #ifndef HEADER
 #define HEADER
 
@@ -166,5 +168,22 @@ void parallel_single() {
 }
 // TERM_DEBUG-DAG: [[DBG_LOC_START]] = !DILocation(line: [[@LINE-12]],
 // TERM_DEBUG-DAG: [[DBG_LOC_END]] = !DILocation(line: [[@LINE-3]],
-
+#endif
+#else
+// ARRAY-LABEL: array_func
+struct St {
+  int a, b;
+  St() : a(0), b(0) {}
+  St &operator=(const St &) { return *this; };
+  ~St() {}
+};
+
+void array_func(int a[3], St s[2]) {
+// ARRAY: call void @__kmpc_copyprivate(%ident_t* @{{.+}}, i32 %{{.+}}, i64 16, i8* %{{.+}}, void (i8*, i8*)* [[CPY:@.+]], i32 %{{.+}})
+#pragma omp single copyprivate(a, s)
+  ;
+}
+// ARRAY: define internal void [[CPY]]
+// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* %{{.+}}, i64 12, i32 4, i1 false)
+// ARRAY: call dereferenceable(8) %struct.St* @_ZN2StaSERKS_(%struct.St* %{{.+}}, %struct.St* dereferenceable(8) %{{.+}})
 #endif

clang/test/OpenMP/task_firstprivate_codegen.cpp

@@ -3,10 +3,12 @@
 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s
 // expected-no-diagnostics
 
 // It doesn't pass on win32.
 // REQUIRES: shell
+#ifndef ARRAY
 #ifndef HEADER
 #define HEADER
 
@@ -397,4 +399,26 @@ int main() {
 // CHECK: ret i32
 
 #endif
+#else
+// ARRAY-LABEL: array_func
+struct St {
+  int a, b;
+  St() : a(0), b(0) {}
+  St(const St &) {}
+  ~St() {}
+};
+
+void array_func(int a[3], St s[2]) {
+// ARRAY: call i8* @__kmpc_omp_task_alloc(
+// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* %{{.+}}, i64 12, i32 4, i1 false)
+// ARRAY: call void @_ZN2StC1ERKS_(%struct.St* %{{.+}}, %struct.St* dereferenceable(8) %{{.+}})
+// ARRAY: store i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** %{{.+}},
+// ARRAY: call i32 @__kmpc_omp_task(
+// ARRAY: define internal i32 [[DESTRUCTORS]](i32,
+// ARRAY: call void @_ZN2StD1Ev(%struct.St* %{{.+}})
+// ARRAY: br i1
+#pragma omp task firstprivate(a, s)
+  ;
+}
+#endif
 

clang/test/OpenMP/task_private_codegen.cpp

@@ -3,11 +3,13 @@
 // RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s
 // RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s
+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s
 // expected-no-diagnostics
 
 // It doesn't pass on win32. Investigating.
 // REQUIRES: shell
 
+#ifndef ARRAY
 #ifndef HEADER
 #define HEADER
 
@@ -360,4 +362,26 @@ int main() {
 // CHECK: ret i32
 
 #endif
+#else
+// ARRAY-LABEL: array_func
+struct St {
+  int a, b;
+  St() : a(0), b(0) {}
+  St &operator=(const St &) { return *this; };
+  ~St() {}
+};
+
+void array_func(int a[2], St s[2]) {
+// ARRAY: call i8* @__kmpc_omp_task_alloc(
+// ARRAY-NOT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* %{{.+}}, i64 12, i32 4, i1 false)
+// ARRAY: call void @_ZN2StC1Ev(%struct.St* %{{.+}})
+// ARRAY: store i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[DESTRUCTORS:@.+]] to i32 (i32, i8*)*), i32 (i32, i8*)** %{{.+}},
+// ARRAY: call i32 @__kmpc_omp_task(
+// ARRAY: define internal i32 [[DESTRUCTORS]](i32,
+// ARRAY: call void @_ZN2StD1Ev(%struct.St* %{{.+}})
+// ARRAY: br i1
+#pragma omp task private(a, s)
+  ;
+}
+#endif