Fix for Bug 30639: CGDebugInfo Null dereference with OpenMP array access

[erichkeane]

Summary

OpenMP creates a variable array type with a a null size-expr. The Debug generation failed to due to this. This patch corrects the openmp implementation, updates the tests, and adds a new one for this condition.

[ABataev]

I think the fix is not quite correct. I believe it's better to replace a call of getVariableArrayDecayedType() in CGStmtOpenMP.cpp by a call to getCanonicalParamType(). Try to replace it and check the result.

[erichkeane]

Andrey-
It seems that getVariableArrayDecayedType and getCanonicalParamType return the same type in this case (at least in the reproduction). I could definitely change the call, though the changes to CGDebugInfo.cpp are apparently also necessary even in that case.

Is there additional logic that should happen in CGStmtOpenMP that I'm missing?

[erichkeane]

Emailed with Alexey who identified the problem in the OpenMP implementation that was resulting in incomplete array types being emitted, which broke an assumption in the new CGDebugInfo code.

[ABataev]

LG with a nit

lib/CodeGen/CGStmtOpenMP.cpp

312

Remove this->

[erichkeane]

Fixed Alexey's nit. Will need someone else to commit this, I don't yet have commit access.

[ABataev]

I will commit it for you, Erich, no problems

lib/CodeGen/CGStmtOpenMP.cpp

if (I->capturesVariable() || I->capturesVariableByCopy()) {
       CapVar = I->getCapturedVar();
       II = CapVar->getIdentifier();
     } else if (I->capturesThis())
       II = &getContext().Idents.get("this");
     else {
       assert(I->capturesVariableArrayType());
       II = &getContext().Idents.get("vla");
     }
-    if (ArgType->isVariablyModifiedType())
-      ArgType = getContext().getVariableArrayDecayedType(ArgType);
+    if (ArgType->isVariablyModifiedType()) {
+      bool IsReference = ArgType->isLValueReferenceType();
+      ArgType =
+          getContext().getCanonicalParamType(ArgType.getNonReferenceType());
+      if (IsReference && !ArgType->isPointerType()) {
+        ArgType = getContext().getLValueReferenceType(
+            ArgType, /*SpelledAsLValue=*/false);
+      }
+    }
     Args.push_back(ImplicitParamDecl::Create(getContext(), nullptr,
                                              FD->getLocation(), II, ArgType));
     ++I;
   }
   Args.append(
       std::next(CD->param_begin(), CD->getContextParamPosition() + 1),
       CD->param_end());
 
if (FD->hasCapturedVLAType()) {
           EmitLoadOfLValue(CastedArgLVal, SourceLocation()).getScalarVal();
       auto VAT = FD->getCapturedVLAType();
       VLASizeMap[VAT->getSizeExpr()] = ExprArg;
     } else if (I->capturesVariable()) {
       auto *Var = I->getCapturedVar();
       QualType VarTy = Var->getType();
       Address ArgAddr = ArgLVal.getAddress();
       if (!VarTy->isReferenceType()) {
+        if (ArgLVal.getType()->isLValueReferenceType()) {
           ArgAddr = EmitLoadOfReference(
               ArgAddr, ArgLVal.getType()->castAs<ReferenceType>());
+        } else {
+          assert(ArgLVal.getType()->isPointerType());
+          ArgAddr = this->EmitLoadOfPointer(
+              ArgAddr, ArgLVal.getType()->castAs<PointerType>());
+        }
       }
       setAddrOfLocalVar(
           Var, Address(ArgAddr.getPointer(), getContext().getDeclAlign(Var)));
     } else if (I->capturesVariableByCopy()) {
       assert(!FD->getType()->isAnyPointerType() &&
              "Not expecting a captured pointer.");
       auto *Var = I->getCapturedVar();
       QualType VarTy = Var->getType();

test/CodeGenCXX/debug-info-openmp-array.cpp

+// RUN: %clang -target x86_64-unknown-unknown -fverbose-asm -fopenmp -g -O0 -S -emit-llvm %s -o - | FileCheck %s
+
+
+void f(int m) {
+  int i;
+  int cen[m];
+#pragma omp parallel for
+  for (i = 0; i < m; ++i) {
+    cen[i] = i;
+  }
+}
+
+// CHECK: !DICompositeType(tag: DW_TAG_array_type,
+// CHECK-NOT: size:
+// CHECK-SAME: align: 32
+// CHECK-SAME:                              elements: [[ELEM_TYPE:![0-9]+]]
+// CHECK: !DISubrange(count: -1)

test/OpenMP/for_reduction_codegen.cpp

 // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]],
 // CHECK: [[T_VAR1_RHS_VAL:%.+]] = load float, float* [[T_VAR1_RHS]],
 // CHECK: [[CMP:%.+]] = fcmp olt float [[T_VAR1_LHS_VAL]], [[T_VAR1_RHS_VAL]]
 // CHECK: br i1 [[CMP]]
 // CHECK: [[UP:%.+]] = phi float
 // CHECK: store float [[UP]], float* [[T_VAR1_LHS]],
 // CHECK: ret void
 
-// CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* nonnull %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}})
+// CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}})
 
 // Reduction list for runtime.
 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
 
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
 // CHECK: [[IDX1:%.+]] = mul nsw i64 1, %{{.+}}
 // CHECK: [[LB1:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]

 // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}})
 // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8*
 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false)
 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
 // CHECK: br i1 [[DONE]],
 
 // CHECK: ret void
 
-// CHECK: define internal void [[MAIN_MICROTASK2]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* nonnull %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}})
+// CHECK: define internal void [[MAIN_MICROTASK2]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(160) %{{.+}})
 
 // CHECK: [[ARRS_PRIV:%.+]] = alloca [10 x [4 x [[S_FLOAT_TY]]]],
 
 // Reduction list for runtime.
 // CHECK: [[RED_LIST:%.+]] = alloca [3 x i8*],
 
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 

test/OpenMP/for_reduction_codegen_UDR.cpp

 
 // var1_lhs = var1_lhs.operator &&(var1_rhs);
 // CHECK: fmul float
 
 // t_var1_lhs = min(t_var1_lhs, t_var1_rhs);
 // CHECK: fadd float 5.550000e+02, %
 // CHECK: ret void
 
-// CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* nonnull %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(480) %{{.+}})
+// CHECK: define internal void [[MAIN_MICROTASK1]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, [2 x i32]* dereferenceable(8) %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(480) %{{.+}})
 
 // Reduction list for runtime.
 // CHECK: [[RED_LIST:%.+]] = alloca [4 x i8*],
 
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 
 // CHECK: [[IDX1:%.+]] = mul nsw i64 1, %{{.+}}
 // CHECK: [[LB1:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]

 // CHECK: br i1 [[ISEMPTY]],
 // CHECK: phi [[S_FLOAT_TY]]*
 // CHECK: call void @_Z3redR6BaseS1RKS_(%
 // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]]
 // CHECK: br i1 [[DONE]],
 
 // CHECK: ret void
 
-// CHECK: define internal void [[MAIN_MICROTASK2]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* nonnull %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(480) %{{.+}})
+// CHECK: define internal void [[MAIN_MICROTASK2]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, i64 %{{.+}}, i64 %{{.+}}, i32* %{{.+}}, [10 x [4 x [[S_FLOAT_TY]]]]* dereferenceable(480) %{{.+}})
 
 // CHECK: [[ARRS_PRIV:%.+]] = alloca [10 x [4 x [[S_FLOAT_TY]]]],
 
 // Reduction list for runtime.
 // CHECK: [[RED_LIST:%.+]] = alloca [3 x i8*],
 
 // CHECK: store i{{[0-9]+}}* [[GTID_ADDR]], i{{[0-9]+}}** [[GTID_ADDR_ADDR:%.+]],
 

test/OpenMP/target_firstprivate_codegen.cpp

#pragma omp target firstprivate(aa,b,bn,c,cn,d)
   
   // CHECK:  [[BASE_PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
   // CHECK:  [[PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
   // CHECK:  [[SIZES_ARG2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[SIZET2]],  i{{[0-9]+}} 0, i{{[0-9]+}} 0
   // CHECK: {{.+}} = call i32 @__tgt_target(i32 -1, {{.+}}, i32 9, i8** [[BASE_PTR_GEP_ARG2]], i8** [[PTR_GEP_ARG2]], i[[SZ]]* [[SIZES_ARG2]], i32* getelementptr inbounds ([9 x i32], [9 x i32]* [[MAPT2]], i32 0, i32 0))
   
   // make sure that firstprivate variables are generated in all cases and that we use those instances for operations inside the
   // target region
-  // TCHECK:  define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A2_IN:%.+]], [10 x float]* {{.+}} [[B_IN:%.+]], i{{[0-9]+}} [[BN_SZ:%.+]], float* {{.+}} [[BN_IN:%.+]], [5 x [10 x double]]* {{.+}} [[C_IN:%.+]], i{{[0-9]+}} [[CN_SZ1:%.+]], i{{[0-9]+}} [[CN_SZ2:%.+]], double* {{.+}} [[CN_IN:%.+]], [[TT]]* {{.+}} [[D_IN:%.+]])
+  // TCHECK:  define {{.*}}void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A2_IN:%.+]], [10 x float]* {{.+}} [[B_IN:%.+]], i{{[0-9]+}} [[BN_SZ:%.+]], float* [[BN_IN:%.+]], [5 x [10 x double]]* {{.+}} [[C_IN:%.+]], i{{[0-9]+}} [[CN_SZ1:%.+]], i{{[0-9]+}} [[CN_SZ2:%.+]], double* [[CN_IN:%.+]], [[TT]]* {{.+}} [[D_IN:%.+]])
   // TCHECK:  [[A2_ADDR:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[B_ADDR:%.+]] = alloca [10 x float]*,
   // TCHECK:  [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[BN_ADDR:%.+]] = alloca float*,
   // TCHECK:  [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*,
   // TCHECK:  [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}},
   // TCHECK:  [[CN_ADDR:%.+]] = alloca double*,

test/OpenMP/target_map_codegen.cpp

void implicit_maps_variable_length_array (int a){
 
   // CK13: call void [[KERNEL:@.+]](i[[sz]] {{.+}}, i[[sz]] {{.+}}, double* [[DECL]])
   #pragma omp target
   {
     vla[1][3] += 1.0;
   }
 }
 
-// CK13: define internal void [[KERNEL]](i[[sz]] [[VLA0:%.+]], i[[sz]] [[VLA1:%.+]], double* {{.+}}[[ARG:%.+]])
+// CK13: define internal void [[KERNEL]](i[[sz]] [[VLA0:%.+]], i[[sz]] [[VLA1:%.+]], double* {{.*}}[[ARG:%.+]])
 // CK13: [[ADDR0:%.+]] = alloca i[[sz]],
 // CK13: [[ADDR1:%.+]] = alloca i[[sz]],
 // CK13: [[ADDR2:%.+]] = alloca double*,
 // CK13: store i[[sz]] [[VLA0]], i[[sz]]* [[ADDR0]],
 // CK13: store i[[sz]] [[VLA1]], i[[sz]]* [[ADDR1]],
 // CK13: store double* [[ARG]], double** [[ADDR2]],
 // CK13: {{.+}} = load i[[sz]],  i[[sz]]* [[ADDR0]],
 // CK13: {{.+}} = load i[[sz]],  i[[sz]]* [[ADDR1]],