Trying to fix undefined symbol error caused by iterator variable #141507

ShashwathiNavada · 2025-05-26T16:59:12Z

When a mapper is declared with an iterator variable inside the map clause, it results in unintended behavior due to the iterator being implicitly created but left uninitialized.
Testcase:

typedef struct myvec{
    size_t len;
    double *data;
} myvec_t;

#pragma omp declare mapper(id:myvec_t v) map( iterator( iterator_variable=0:v.len), tofrom: v.data[iterator_variable]) 

int main() 
{ 
    int errors = 0;
   myvec_t s;
   #pragma omp target map(mapper(id), to:s)
     {
     }
  return 0;
}

The error we get while compiling this is:

/usr/lib64/gcc/x86_64-suse-linux/14/../../../../x86_64-suse-linux/bin/ld: /tmp/test-f70647.o: in function `.omp_mapper._ZTS5myvec.id':
test.cpp:(.text+0x21a): undefined reference to `iterator_variable'
/llvm-project/install/bin/clang-linker-wrapper: error: 'ld' failed
clang++: error: linker command failed with exit code 1 (use -v to see invocation)

This patch tries to fix this by initializing the iterator variable to a null constant.

llvmbot · 2025-05-26T16:59:45Z

@llvm/pr-subscribers-flang-openmp

@llvm/pr-subscribers-clang

Author: None (ShashwathiNavada)

Changes

When a mapper is declared with an iterator variable inside the map clause, it results in unintended behavior due to the iterator being implicitly created but left uninitialized.
Testcase:

typedef struct myvec{
    size_t len;
    double *data;
} myvec_t;

#pragma omp declare mapper(id:myvec_t v) map( iterator( iterator_variable=0:v.len), tofrom: v.data[iterator_variable]) 

int main() 
{ 
    int errors = 0;
   myvec_t s;
   #pragma omp target map(mapper(id), to:s)
     {
     }
  return 0;
}

The error we get while compiling this is:

/usr/lib64/gcc/x86_64-suse-linux/14/../../../../x86_64-suse-linux/bin/ld: /tmp/test-f70647.o: in function `.omp_mapper._ZTS5myvec.id':
test.cpp:(.text+0x21a): undefined reference to `iterator_variable'
/llvm-project/install/bin/clang-linker-wrapper: error: 'ld' failed
clang++: error: linker command failed with exit code 1 (use -v to see invocation)

This patch tries to fix this by initializing the iterator variable to a null constant.

Full diff: https://github.com/llvm/llvm-project/pull/141507.diff

4 Files Affected:

(modified) clang/include/clang/Basic/Attr.td (+7)
(modified) clang/lib/CodeGen/CGExpr.cpp (+5)
(modified) clang/lib/Sema/SemaOpenMP.cpp (+2)
(modified) clang/test/OpenMP/declare_mapper_codegen.cpp (+102)

diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index ccd13a4cca4dd..cc629443e9b2d 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -4461,6 +4461,13 @@ def OMPCaptureKind : Attr {
   }];
 }
 
+def OMPIterator : Attr {
+  // This attribute has no spellings as it is only ever created implicitly.
+  let Spellings = [];
+  let SemaHandler = 0;
+  let Documentation = [InternalOnly];
+}
+
 def OMPReferencedVar : Attr {
   // This attribute has no spellings as it is only ever created implicitly.
   let Spellings = [];
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index ec01c87c13b1d..994327f690689 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -2969,6 +2969,11 @@ static LValue EmitGlobalVarDeclLValue(CodeGenFunction &CGF,
 
   llvm::Value *V = CGF.CGM.GetAddrOfGlobalVar(VD);
 
+  if (VD->hasAttr<OMPIteratorAttr>()) {
+  llvm::GlobalVariable *Var = llvm::dyn_cast<llvm::GlobalVariable>(V);
+  Var->setInitializer(CGF.CGM.EmitNullConstant(E->getType()));
+  }
+
   if (VD->getTLSKind() != VarDecl::TLS_None)
     V = CGF.Builder.CreateThreadLocalAddress(V);
 
diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index f16f841d62edd..1756c20ce486c 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -24476,6 +24476,8 @@ ExprResult SemaOpenMP::ActOnOMPIteratorExpr(Scope *S,
         VarDecl::Create(Context, SemaRef.CurContext, StartLoc, D.DeclIdentLoc,
                         D.DeclIdent, DeclTy, TInfo, SC_None);
     VD->setImplicit();
+    VD->addAttr(OMPIteratorAttr::CreateImplicit(
+        Context, SourceRange(StartLoc)));
     if (S) {
       // Check for conflicting previous declaration.
       DeclarationNameInfo NameInfo(VD->getDeclName(), D.DeclIdentLoc);
diff --git a/clang/test/OpenMP/declare_mapper_codegen.cpp b/clang/test/OpenMP/declare_mapper_codegen.cpp
index 81453223b2a27..7dc32d0ae12ff 100644
--- a/clang/test/OpenMP/declare_mapper_codegen.cpp
+++ b/clang/test/OpenMP/declare_mapper_codegen.cpp
@@ -1055,4 +1055,106 @@ void foo(int a){
 
 #endif // CK4
 
+///==========================================================================///
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix CK5 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-version=52 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix CK5 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-version=52 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+
+#ifdef CK5
+typedef struct myvec {
+    int a;
+    double *b;
+} myvec_t;
+
+#pragma omp declare mapper(id: myvec_t v) map(iterator(it=0:v.a), tofrom: v.b[it])
+// CK5: @[[ITER:[a-zA-Z0-9_]+]] = global i32 0, align 4
+
+void foo(){ 
+  myvec_t s;
+  #pragma omp target map(mapper(id), to:s)
+  {
+  }
+}
+
+// CK5: define {{.*}}void [[MPRFUNC:@[.]omp_mapper[.].*myvec[.]id]](ptr noundef [[HANDLE:%.+]], ptr noundef [[BPTR:%.+]], ptr noundef [[BEGIN:%.+]], i64 noundef [[BYTESIZE:%.+]], i64 noundef [[TYPE:%.+]], ptr{{.*}})
+// CK5-DAG: [[SIZE:%.+]] = udiv exact i64 [[BYTESIZE]], {{.*}}
+// CK5-DAG: [[PTREND:%.+]] = getelementptr %struct.myvec, ptr [[BEGIN]], i64 [[SIZE]]
+// CK5-DAG: [[ISARRAY:%.+]] = icmp sgt i64 [[SIZE]], 1
+// CK5-DAG: [[PTRSNE:%.+]] = icmp ne ptr [[BPTR]], [[BEGIN]]
+// CK5-DAG: [[PTRANDOBJ:%.+]] = and i64 [[TYPE]], 16
+// CK5-DAG: [[ISPTRANDOBJ:%.+]] = icmp ne i64 [[PTRANDOBJ]], 0
+// CK5-DAG: [[CMPA:%.+]] = and i1 [[PTRSNE]], [[ISPTRANDOBJ]]
+// CK5-DAG: [[CMP:%.+]] = or i1 [[ISARRAY]], [[CMPA]]
+// CK5-DAG: [[TYPEDEL:%.+]] = and i64 [[TYPE]], 8
+// CK5-DAG: [[ISNOTDEL:%.+]] = icmp eq i64 [[TYPEDEL]], 0
+// CK5-DAG: [[CMP1:%.+]] = and i1 [[CMP]], [[ISNOTDEL]]
+// CK5: br i1 [[CMP1]], label %[[INITEVALDEL:[^,]+]], label %[[LHEAD:[^,]+]]
+
+// CK5: [[INITEVALDEL]]
+// CK5-DAG: [[ARRSIZE:%.+]] = mul nuw i64 [[SIZE]], {{.*}}
+
+// Remove movement mappings and mark as implicit
+// CK5-DAG: [[ITYPE:%.+]] = and i64 [[TYPE]], -4
+// CK5-DAG: [[ITYPE1:%.+]] = or i64 [[ITYPE]], 512
+// CK5: call void @__tgt_push_mapper_component(ptr [[HANDLE]], ptr [[BPTR]], ptr [[BEGIN]], i64 [[ARRSIZE]], i64 [[ITYPE1]], {{.*}})
+// CK5: br label %[[LHEAD:[^,]+]]
+
+// CK5: [[LHEAD]]
+// CK5: [[ISEMPTY:%.+]] = icmp eq ptr [[BEGIN]], [[PTREND]]
+// CK5: br i1 [[ISEMPTY]], label %[[DONE:[^,]+]], label %[[LBODY:[^,]+]]
+// CK5: [[LBODY]]
+// CK5: [[PTR:%.+]] = phi ptr [ [[BEGIN]], %{{.+}} ], [ [[PTRNEXT:%.+]], %[[LCORRECT:[^,]+]] ]
+// CK5-DAG: [[ABEGIN:%.+]] = getelementptr inbounds nuw %struct.myvec, ptr [[PTR]], i32 0, i32 1
+// CK5-DAG: load i32, ptr @[[ITER]], align 4
+// CK5-DAG: [[PRESIZE:%.+]] = call i64 @__tgt_mapper_num_components(ptr [[HANDLE]])
+// CK5-DAG: [[SHIPRESIZE:%.+]] = shl i64 [[PRESIZE]], 48
+// CK5-DAG: [[MEMBERTYPE:%.+]] = add nuw i64 0, [[SHIPRESIZE]]
+// CK5-DAG: [[TYPETF:%.+]] = and i64 [[TYPE]], 3
+// CK5-DAG: [[ISALLOC:%.+]] = icmp eq i64 [[TYPETF]], 0
+// CK5-DAG: br i1 [[ISALLOC]], label %[[ALLOC:[^,]+]], label %[[ALLOCELSE:[^,]+]]
+// CK5-DAG: [[ALLOC]]
+// CK5-DAG: [[ALLOCTYPE:%.+]] = and i64 [[MEMBERTYPE]], -4
+// CK5-DAG: br label %[[TYEND:[^,]+]]
+// CK5-DAG: [[ALLOCELSE]]
+// CK5-DAG: [[ISTO:%.+]] = icmp eq i64 [[TYPETF]], 1
+// CK5-DAG: br i1 [[ISTO]], label %[[TO:[^,]+]], label %[[TOELSE:[^,]+]]
+// CK5-DAG: [[TO]]
+// CK5-DAG: [[TOTYPE:%.+]] = and i64 [[MEMBERTYPE]], -3
+// CK5-DAG: br label %[[TYEND]]
+// CK5-DAG: [[TOELSE]]
+// CK5-DAG: [[ISFROM:%.+]] = icmp eq i64 [[TYPETF]], 2
+// CK5-DAG: br i1 [[ISFROM]], label %[[FROM:[^,]+]], label %[[TYEND]]
+// CK5-DAG: [[FROM]]
+// CK5-DAG: [[FROMTYPE:%.+]] = and i64 [[MEMBERTYPE]], -2
+// CK5-DAG: br label %[[TYEND]]
+// CK5-DAG: [[TYEND]]
+// CK5-DAG: [[TYPE1:%.+]] = phi i64 [ [[ALLOCTYPE]], %[[ALLOC]] ], [ [[TOTYPE]], %[[TO]] ], [ [[FROMTYPE]], %[[FROM]] ], [ [[MEMBERTYPE]], %[[TOELSE]] ]
+// CK5: call void @__tgt_push_mapper_component(ptr [[HANDLE]], ptr [[PTR]], ptr [[ABEGIN]], i64 {{.*}}, i64 [[TYPE1]], {{.*}})
+// CK5: [[PTRNEXT]] = getelementptr %struct.myvec, ptr [[PTR]], i32 1
+// CK5: [[ISDONE:%.+]] = icmp eq ptr [[PTRNEXT]], [[PTREND]]
+// CK5: br i1 [[ISDONE]], label %[[LEXIT:[^,]+]], label %[[LBODY]]
+
+// CK5: [[LEXIT]]
+// CK5: [[ISARRAY:%.+]] = icmp sgt i64 [[SIZE]], 1
+// CK5: [[TYPEDEL:%.+]] = and i64 [[TYPE]], 8
+// CK5: [[ISNOTDEL:%.+]] = icmp ne i64 [[TYPEDEL]], {{.*}}
+// CK5: [[CMP1:%.+]] = and i1 [[ISARRAY]], [[ISNOTDEL]]
+// CK5-DAG: [[ARRSIZE:%.+]] = mul nuw i64 [[SIZE]], {{.*}}
+
+// Remove movement mappings and mark as implicit
+// CK5-DAG: [[DTYPE:%.+]] = and i64 [[TYPE]], -4
+// CK5-DAG: [[DTYPE1:%.+]] = or i64 [[DTYPE]], 512
+// CK5: call void @__tgt_push_mapper_component(ptr [[HANDLE]], ptr [[BPTR]], ptr [[BEGIN]], i64 [[ARRSIZE]], i64 [[DTYPE1]], {{.*}})
+// CK5: br label %[[DONE]]
+// CK5: [[DONE]]
+// CK5: ret void
+
+#endif // CK5
+
 #endif // HEADER

llvmbot · 2025-05-26T16:59:45Z

@llvm/pr-subscribers-clang-codegen

Author: None (ShashwathiNavada)

Changes

When a mapper is declared with an iterator variable inside the map clause, it results in unintended behavior due to the iterator being implicitly created but left uninitialized.
Testcase:

typedef struct myvec{
    size_t len;
    double *data;
} myvec_t;

#pragma omp declare mapper(id:myvec_t v) map( iterator( iterator_variable=0:v.len), tofrom: v.data[iterator_variable]) 

int main() 
{ 
    int errors = 0;
   myvec_t s;
   #pragma omp target map(mapper(id), to:s)
     {
     }
  return 0;
}

The error we get while compiling this is:

/usr/lib64/gcc/x86_64-suse-linux/14/../../../../x86_64-suse-linux/bin/ld: /tmp/test-f70647.o: in function `.omp_mapper._ZTS5myvec.id':
test.cpp:(.text+0x21a): undefined reference to `iterator_variable'
/llvm-project/install/bin/clang-linker-wrapper: error: 'ld' failed
clang++: error: linker command failed with exit code 1 (use -v to see invocation)

This patch tries to fix this by initializing the iterator variable to a null constant.

Full diff: https://github.com/llvm/llvm-project/pull/141507.diff

4 Files Affected:

(modified) clang/include/clang/Basic/Attr.td (+7)
(modified) clang/lib/CodeGen/CGExpr.cpp (+5)
(modified) clang/lib/Sema/SemaOpenMP.cpp (+2)
(modified) clang/test/OpenMP/declare_mapper_codegen.cpp (+102)

diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index ccd13a4cca4dd..cc629443e9b2d 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -4461,6 +4461,13 @@ def OMPCaptureKind : Attr {
   }];
 }
 
+def OMPIterator : Attr {
+  // This attribute has no spellings as it is only ever created implicitly.
+  let Spellings = [];
+  let SemaHandler = 0;
+  let Documentation = [InternalOnly];
+}
+
 def OMPReferencedVar : Attr {
   // This attribute has no spellings as it is only ever created implicitly.
   let Spellings = [];
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index ec01c87c13b1d..994327f690689 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -2969,6 +2969,11 @@ static LValue EmitGlobalVarDeclLValue(CodeGenFunction &CGF,
 
   llvm::Value *V = CGF.CGM.GetAddrOfGlobalVar(VD);
 
+  if (VD->hasAttr<OMPIteratorAttr>()) {
+  llvm::GlobalVariable *Var = llvm::dyn_cast<llvm::GlobalVariable>(V);
+  Var->setInitializer(CGF.CGM.EmitNullConstant(E->getType()));
+  }
+
   if (VD->getTLSKind() != VarDecl::TLS_None)
     V = CGF.Builder.CreateThreadLocalAddress(V);
 
diff --git a/clang/lib/Sema/SemaOpenMP.cpp b/clang/lib/Sema/SemaOpenMP.cpp
index f16f841d62edd..1756c20ce486c 100644
--- a/clang/lib/Sema/SemaOpenMP.cpp
+++ b/clang/lib/Sema/SemaOpenMP.cpp
@@ -24476,6 +24476,8 @@ ExprResult SemaOpenMP::ActOnOMPIteratorExpr(Scope *S,
         VarDecl::Create(Context, SemaRef.CurContext, StartLoc, D.DeclIdentLoc,
                         D.DeclIdent, DeclTy, TInfo, SC_None);
     VD->setImplicit();
+    VD->addAttr(OMPIteratorAttr::CreateImplicit(
+        Context, SourceRange(StartLoc)));
     if (S) {
       // Check for conflicting previous declaration.
       DeclarationNameInfo NameInfo(VD->getDeclName(), D.DeclIdentLoc);
diff --git a/clang/test/OpenMP/declare_mapper_codegen.cpp b/clang/test/OpenMP/declare_mapper_codegen.cpp
index 81453223b2a27..7dc32d0ae12ff 100644
--- a/clang/test/OpenMP/declare_mapper_codegen.cpp
+++ b/clang/test/OpenMP/declare_mapper_codegen.cpp
@@ -1055,4 +1055,106 @@ void foo(int a){
 
 #endif // CK4
 
+///==========================================================================///
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix CK5 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-version=52 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix CK5 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-version=52 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm -femit-all-decls -disable-llvm-passes %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-version=52 -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -femit-all-decls -disable-llvm-passes -o %t %s
+
+#ifdef CK5
+typedef struct myvec {
+    int a;
+    double *b;
+} myvec_t;
+
+#pragma omp declare mapper(id: myvec_t v) map(iterator(it=0:v.a), tofrom: v.b[it])
+// CK5: @[[ITER:[a-zA-Z0-9_]+]] = global i32 0, align 4
+
+void foo(){ 
+  myvec_t s;
+  #pragma omp target map(mapper(id), to:s)
+  {
+  }
+}
+
+// CK5: define {{.*}}void [[MPRFUNC:@[.]omp_mapper[.].*myvec[.]id]](ptr noundef [[HANDLE:%.+]], ptr noundef [[BPTR:%.+]], ptr noundef [[BEGIN:%.+]], i64 noundef [[BYTESIZE:%.+]], i64 noundef [[TYPE:%.+]], ptr{{.*}})
+// CK5-DAG: [[SIZE:%.+]] = udiv exact i64 [[BYTESIZE]], {{.*}}
+// CK5-DAG: [[PTREND:%.+]] = getelementptr %struct.myvec, ptr [[BEGIN]], i64 [[SIZE]]
+// CK5-DAG: [[ISARRAY:%.+]] = icmp sgt i64 [[SIZE]], 1
+// CK5-DAG: [[PTRSNE:%.+]] = icmp ne ptr [[BPTR]], [[BEGIN]]
+// CK5-DAG: [[PTRANDOBJ:%.+]] = and i64 [[TYPE]], 16
+// CK5-DAG: [[ISPTRANDOBJ:%.+]] = icmp ne i64 [[PTRANDOBJ]], 0
+// CK5-DAG: [[CMPA:%.+]] = and i1 [[PTRSNE]], [[ISPTRANDOBJ]]
+// CK5-DAG: [[CMP:%.+]] = or i1 [[ISARRAY]], [[CMPA]]
+// CK5-DAG: [[TYPEDEL:%.+]] = and i64 [[TYPE]], 8
+// CK5-DAG: [[ISNOTDEL:%.+]] = icmp eq i64 [[TYPEDEL]], 0
+// CK5-DAG: [[CMP1:%.+]] = and i1 [[CMP]], [[ISNOTDEL]]
+// CK5: br i1 [[CMP1]], label %[[INITEVALDEL:[^,]+]], label %[[LHEAD:[^,]+]]
+
+// CK5: [[INITEVALDEL]]
+// CK5-DAG: [[ARRSIZE:%.+]] = mul nuw i64 [[SIZE]], {{.*}}
+
+// Remove movement mappings and mark as implicit
+// CK5-DAG: [[ITYPE:%.+]] = and i64 [[TYPE]], -4
+// CK5-DAG: [[ITYPE1:%.+]] = or i64 [[ITYPE]], 512
+// CK5: call void @__tgt_push_mapper_component(ptr [[HANDLE]], ptr [[BPTR]], ptr [[BEGIN]], i64 [[ARRSIZE]], i64 [[ITYPE1]], {{.*}})
+// CK5: br label %[[LHEAD:[^,]+]]
+
+// CK5: [[LHEAD]]
+// CK5: [[ISEMPTY:%.+]] = icmp eq ptr [[BEGIN]], [[PTREND]]
+// CK5: br i1 [[ISEMPTY]], label %[[DONE:[^,]+]], label %[[LBODY:[^,]+]]
+// CK5: [[LBODY]]
+// CK5: [[PTR:%.+]] = phi ptr [ [[BEGIN]], %{{.+}} ], [ [[PTRNEXT:%.+]], %[[LCORRECT:[^,]+]] ]
+// CK5-DAG: [[ABEGIN:%.+]] = getelementptr inbounds nuw %struct.myvec, ptr [[PTR]], i32 0, i32 1
+// CK5-DAG: load i32, ptr @[[ITER]], align 4
+// CK5-DAG: [[PRESIZE:%.+]] = call i64 @__tgt_mapper_num_components(ptr [[HANDLE]])
+// CK5-DAG: [[SHIPRESIZE:%.+]] = shl i64 [[PRESIZE]], 48
+// CK5-DAG: [[MEMBERTYPE:%.+]] = add nuw i64 0, [[SHIPRESIZE]]
+// CK5-DAG: [[TYPETF:%.+]] = and i64 [[TYPE]], 3
+// CK5-DAG: [[ISALLOC:%.+]] = icmp eq i64 [[TYPETF]], 0
+// CK5-DAG: br i1 [[ISALLOC]], label %[[ALLOC:[^,]+]], label %[[ALLOCELSE:[^,]+]]
+// CK5-DAG: [[ALLOC]]
+// CK5-DAG: [[ALLOCTYPE:%.+]] = and i64 [[MEMBERTYPE]], -4
+// CK5-DAG: br label %[[TYEND:[^,]+]]
+// CK5-DAG: [[ALLOCELSE]]
+// CK5-DAG: [[ISTO:%.+]] = icmp eq i64 [[TYPETF]], 1
+// CK5-DAG: br i1 [[ISTO]], label %[[TO:[^,]+]], label %[[TOELSE:[^,]+]]
+// CK5-DAG: [[TO]]
+// CK5-DAG: [[TOTYPE:%.+]] = and i64 [[MEMBERTYPE]], -3
+// CK5-DAG: br label %[[TYEND]]
+// CK5-DAG: [[TOELSE]]
+// CK5-DAG: [[ISFROM:%.+]] = icmp eq i64 [[TYPETF]], 2
+// CK5-DAG: br i1 [[ISFROM]], label %[[FROM:[^,]+]], label %[[TYEND]]
+// CK5-DAG: [[FROM]]
+// CK5-DAG: [[FROMTYPE:%.+]] = and i64 [[MEMBERTYPE]], -2
+// CK5-DAG: br label %[[TYEND]]
+// CK5-DAG: [[TYEND]]
+// CK5-DAG: [[TYPE1:%.+]] = phi i64 [ [[ALLOCTYPE]], %[[ALLOC]] ], [ [[TOTYPE]], %[[TO]] ], [ [[FROMTYPE]], %[[FROM]] ], [ [[MEMBERTYPE]], %[[TOELSE]] ]
+// CK5: call void @__tgt_push_mapper_component(ptr [[HANDLE]], ptr [[PTR]], ptr [[ABEGIN]], i64 {{.*}}, i64 [[TYPE1]], {{.*}})
+// CK5: [[PTRNEXT]] = getelementptr %struct.myvec, ptr [[PTR]], i32 1
+// CK5: [[ISDONE:%.+]] = icmp eq ptr [[PTRNEXT]], [[PTREND]]
+// CK5: br i1 [[ISDONE]], label %[[LEXIT:[^,]+]], label %[[LBODY]]
+
+// CK5: [[LEXIT]]
+// CK5: [[ISARRAY:%.+]] = icmp sgt i64 [[SIZE]], 1
+// CK5: [[TYPEDEL:%.+]] = and i64 [[TYPE]], 8
+// CK5: [[ISNOTDEL:%.+]] = icmp ne i64 [[TYPEDEL]], {{.*}}
+// CK5: [[CMP1:%.+]] = and i1 [[ISARRAY]], [[ISNOTDEL]]
+// CK5-DAG: [[ARRSIZE:%.+]] = mul nuw i64 [[SIZE]], {{.*}}
+
+// Remove movement mappings and mark as implicit
+// CK5-DAG: [[DTYPE:%.+]] = and i64 [[TYPE]], -4
+// CK5-DAG: [[DTYPE1:%.+]] = or i64 [[DTYPE]], 512
+// CK5: call void @__tgt_push_mapper_component(ptr [[HANDLE]], ptr [[BPTR]], ptr [[BEGIN]], i64 [[ARRSIZE]], i64 [[DTYPE1]], {{.*}})
+// CK5: br label %[[DONE]]
+// CK5: [[DONE]]
+// CK5: ret void
+
+#endif // CK5
+
 #endif // HEADER

github-actions · 2025-05-26T17:01:27Z

✅ With the latest revision this PR passed the C/C++ code formatter.

efriedma-quic · 2025-05-27T17:08:38Z

clang/lib/CodeGen/CGExpr.cpp

@@ -2969,6 +2969,11 @@ static LValue EmitGlobalVarDeclLValue(CodeGenFunction &CGF,

  llvm::Value *V = CGF.CGM.GetAddrOfGlobalVar(VD);

+  if (VD->hasAttr<OMPIteratorAttr>()) {
+    llvm::GlobalVariable *Var = llvm::dyn_cast<llvm::GlobalVariable>(V);
+    Var->setInitializer(CGF.CGM.EmitNullConstant(E->getType()));


This is not the place to set an initializer for a variable.

For most variable definitions, there's a single location where it makes sense to "define" the variable.

This is not the place to set an initializer for a variable.

For most variable definitions, there's a single location where it makes sense to "define" the variable.

Thank you for the response, I have moved the definition. Does it look good now?

efriedma-quic

To expand on what I mean...

For variables which have a strong definition, we usually don't emit them lazily: there's some statement or declaration which is the "definition" point, and we emit the definition as part of visiting that statement or declaration. This ensures consistency: we consistently emit a definition for things that need a definition.

For variables where the definition isn't necessarily needed, we also have a mechanism for lazily defining them; see DeferredDecls.

I don't want to invent a new way of emitting variable definitions specifically for OpenMP.

…EmitGlobal function

ShashwathiNavada · 2025-07-23T09:19:45Z

To expand on what I mean...

For variables which have a strong definition, we usually don't emit them lazily: there's some statement or declaration which is the "definition" point, and we emit the definition as part of visiting that statement or declaration. This ensures consistency: we consistently emit a definition for things that need a definition.

For variables where the definition isn't necessarily needed, we also have a mechanism for lazily defining them; see DeferredDecls.

I don't want to invent a new way of emitting variable definitions specifically for OpenMP.

@efriedma-quic Thank you! I've updated the patch to emit iterator variable using the existing global emission logic, similar to how it's done for declare target deferred variables. Let me know if this looks good.

efriedma-quic

That's maybe better?

I spent a little time looking at the testcase, and I saw something weird in the AST: there's a VarDecl at the top level, but it never gets sent down the normal path for top-level decls. I suspect there's a missing call to HandleTopLevelDecl somewhere. That would be the best way to handle this, if possible.

zahiraam · 2025-07-25T12:52:50Z

The reproducer is passing with clang. See https://godbolt.org/z/adnjMndGq

efriedma-quic · 2025-07-25T20:06:01Z

The reproducer is passing with clang. See https://godbolt.org/z/adnjMndGq

The testcase is using -femit-all-decls. I assumed that was just to avoid including some irrelevant bits in the testcase. I guess there could be some interaction with -femit-all-decls specifically, though.

ShashwathiNavada · 2025-07-29T09:11:30Z

The reproducer is passing with clang. See https://godbolt.org/z/adnjMndGq

Since parsing and semantic support for the iterator is in place, the code compiles fine without offloading. However, the issue surfaces only when offloading is actually performed.

ShashwathiNavada · 2025-08-01T08:15:40Z

That's maybe better?

I spent a little time looking at the testcase, and I saw something weird in the AST: there's a VarDecl at the top level, but it never gets sent down the normal path for top-level decls. I suspect there's a missing call to HandleTopLevelDecl somewhere. That would be the best way to handle this, if possible.

@efriedma-quic Thank you for the suggestion,.I have added the call. Could you please take a look and let me know if it looks good?

efriedma-quic

This looks basically right. But I'm not sure all the conditions for calling it are correct. Is getDeclareMapperVarRef relevant? Does it matter what scope we're inserting the variable into? (Can the variable be inserted in a namespace? Can it be a local variable?)

ShashwathiNavada · 2025-08-04T13:46:25Z

This looks basically right. But I'm not sure all the conditions for calling it are correct. Is getDeclareMapperVarRef relevant? Does it matter what scope we're inserting the variable into? (Can the variable be inserted in a namespace? Can it be a local variable?)

@efriedma-quic I've added the condition to check for local variables as you suggested. However, I believe getDeclareMapperVarRef() is relevant in this context, since I'm handling iterators declared in #pragma omp declare mapper. Let me know if you think any further changes are needed.

efriedma-quic

The reason I'm suspicious of the getDeclareMapperVarRef check is that SemaOpenMP::ActOnOMPIteratorExpr creates the variable (VarDecl::Create) regardless. So what's supposed to happen for variables where getDeclareMapperVarRef is false?

Independent of your patch, the following seems to crash. Is it supposed to be valid? Is there an open bug?

typedef struct myvec {
    int a;
    double *b;
} myvec_t;
void foo() {
#pragma omp declare mapper(id: myvec_t v) map(iterator(it=0:v.a), tofrom: v.b[it])
  myvec_t s;
  #pragma omp target map(mapper(id), to:s)
  {
  }
}

ShashwathiNavada · 2025-08-05T17:39:21Z

Independent of your patch, the following seems to crash. Is it supposed to be valid? Is there an open bug?
typedef struct myvec {
    int a;
    double *b;
} myvec_t;
void foo() {
#pragma omp declare mapper(id: myvec_t v) map(iterator(it=0:v.a), tofrom: v.b[it])
  myvec_t s;
  #pragma omp target map(mapper(id), to:s)
  {
  }
}

I think the issue here is the missing memory allocation for s in the testcase. So if we allocate memory for s.b the program works fine.

efriedma-quic · 2025-08-07T20:54:09Z

Still waiting for an answer:

The reason I'm suspicious of the getDeclareMapperVarRef check is that SemaOpenMP::ActOnOMPIteratorExpr creates the variable (VarDecl::Create) regardless. So what's supposed to happen for variables where getDeclareMapperVarRef is false?

ShashwathiNavada · 2025-08-12T14:55:41Z

Still waiting for an answer:

The reason I'm suspicious of the getDeclareMapperVarRef check is that SemaOpenMP::ActOnOMPIteratorExpr creates the variable (VarDecl::Create) regardless. So what's supposed to happen for variables where getDeclareMapperVarRef is false?

This check was added when initially supporting iterators in declare mapper. Although ActOnOMPIteratorExpr always creates the iterator variable, this part is specifically to handle processing that variable for declare mapper purposes. So as far as I understand, even if getDeclareMapperVarRef() is false, the variable exists, but we skip special declare-mapper-related handling here. Please let me know if I’m wrong.

efriedma-quic · 2025-08-12T16:47:16Z

We need to call HandleTopLevelDecl for every translation-unit-level variable we create. ActOnOMPIteratorExpr creates a translation-unit-level variable for every top-level iterator expression. But we're only calling HandleTopLevelDecl for some of them: specifically, the ones where getDeclareMapperVarRef() is true. What happens to the other ones?

ShashwathiNavada · 2025-08-18T15:31:06Z

We need to call HandleTopLevelDecl for every translation-unit-level variable we create. ActOnOMPIteratorExpr creates a translation-unit-level variable for every top-level iterator expression. But we're only calling HandleTopLevelDecl for some of them: specifically, the ones where getDeclareMapperVarRef() is true. What happens to the other ones?

I see your point. I noticed that ActOnOMPIteratorExpr gets called for affinity, depend, and map clauses. Out of these, only map seems to appear at global scope. That probably explains why we don’t call HandleTopLevelDecl here. How do you think we should proceed from here?

efriedma-quic · 2025-08-18T17:12:40Z

If you're convinced it can't happen, add an appropriate assertion.

ShashwathiNavada · 2025-08-19T13:01:25Z

If you're convinced it can't happen, add an appropriate assertion.

Done! please let me know if this looks good or if there is anything else that needs to be done.

Shashwathi N added 2 commits May 19, 2025 14:29

Trying to fix undefined symbol error caused by iterator variable

435206a

Added testcase

Loading
Loading status checks…

00b30ed

llvmbot added clang clang:frontend clang:codegen clang:openmp labels May 26, 2025

Minor changes

Loading
Loading status checks…

f9029e8

efriedma-quic reviewed May 27, 2025

View reviewed changes

Changes

Loading
Loading status checks…

dbdbe3c

llvmbot added the flang:openmp label May 28, 2025

Format

Loading
Loading status checks…

e65ff6b

efriedma-quic reviewed May 28, 2025

View reviewed changes

Shashwathi N added 2 commits July 23, 2025 03:12

Instead of initializing variable manually, emit them as global using …

Loading
Loading status checks…

5c75517

…EmitGlobal function

Format changes

Loading
Loading status checks…

7bbb30a

ShashwathiNavada requested a review from efriedma-quic July 23, 2025 09:50

efriedma-quic reviewed Jul 23, 2025

View reviewed changes

shafik requested a review from zahiraam July 24, 2025 22:57

Shashwathi N and others added 2 commits August 1, 2025 03:11

Added call for HandleTopLevelDecl in SemaOpenmp

Loading
Loading status checks…

bb48a76

Merge branch 'main' into Iterator_fix

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.

GPG key ID: B5690EEEBB952194

Verified
Learn about vigilant mode

Loading
Loading status checks…

f908d2c

ShashwathiNavada added 3 commits August 1, 2025 13:46

Merge branch 'main' into Iterator_fix

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.

GPG key ID: B5690EEEBB952194

Verified
Learn about vigilant mode

Loading
Loading status checks…

224a8e0

Update SemaOpenMP.cpp

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.

GPG key ID: B5690EEEBB952194

Verified
Learn about vigilant mode

Loading
Loading status checks…

731d27b

Merge branch 'main' into Iterator_fix

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.

GPG key ID: B5690EEEBB952194

Verified
Learn about vigilant mode

Loading
Loading status checks…

7ca8c69

Update SemaOpenMP.cpp

Verified

This commit was created on GitHub.com and signed with GitHub’s verified signature.

GPG key ID: B5690EEEBB952194

Verified
Learn about vigilant mode

Loading
Loading status checks…

1e9fccc

efriedma-quic reviewed Aug 1, 2025

View reviewed changes

Added conditions

Loading
Loading status checks…

ffa6239

efriedma-quic reviewed Aug 4, 2025

View reviewed changes

ShashwathiNavada requested a review from efriedma-quic August 7, 2025 05:29

added asertion

Loading
Loading status checks…

8d82ac3

update

Loading
Loading status checks…

dd9267c

Trying to fix undefined symbol error caused by iterator variable #141507

Are you sure you want to change the base?

Trying to fix undefined symbol error caused by iterator variable #141507

Conversation

ShashwathiNavada commented May 26, 2025

Uh oh!

Uh oh!

llvmbot commented May 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented May 26, 2025

Uh oh!

github-actions bot commented May 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

efriedma-quic May 27, 2025

Choose a reason for hiding this comment

Uh oh!

ShashwathiNavada May 28, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

ShashwathiNavada commented Jul 23, 2025

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

zahiraam commented Jul 25, 2025

Uh oh!

efriedma-quic commented Jul 25, 2025

Uh oh!

ShashwathiNavada commented Jul 29, 2025

Uh oh!

Uh oh!

Uh oh!

ShashwathiNavada commented Aug 1, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ShashwathiNavada commented Aug 4, 2025

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

ShashwathiNavada commented Aug 5, 2025

Uh oh!

efriedma-quic commented Aug 7, 2025

Uh oh!

ShashwathiNavada commented Aug 12, 2025

Uh oh!

efriedma-quic commented Aug 12, 2025

Uh oh!

ShashwathiNavada commented Aug 18, 2025

Uh oh!

efriedma-quic commented Aug 18, 2025

Uh oh!

Uh oh!

ShashwathiNavada commented Aug 19, 2025

Uh oh!

Uh oh!

llvmbot commented May 26, 2025 •

edited

Loading

github-actions bot commented May 26, 2025 •

edited

Loading