[clang] [OpenMP] Codegen support for threadset

Ritanya-B-Bharadwaj · Ritanya-B-Bharadwaj · commit d700caad78a8 · 2025-05-07T11:42:19.000-05:00
diff --git a/clang/docs/OpenMPSupport.rst b/clang/docs/OpenMPSupport.rst
@@ -364,7 +364,7 @@ implementation.
 +=============================================================+===========================+===========================+==========================================================================+
 | free-agent threads                                          | :none:`unclaimed`         | :none:`unclaimed`         |                                                                          |
 +-------------------------------------------------------------+---------------------------+---------------------------+--------------------------------------------------------------------------+
-| threadset clause                                            | :`worked on`              | :none:`unclaimed`         |                                                                          |
+| threadset clause                                            | :good:`done`              | :none:`unclaimed`         | https://github.com/llvm/llvm-project/pull/135807                         |
 +-------------------------------------------------------------+---------------------------+---------------------------+--------------------------------------------------------------------------+
 | Recording of task graphs                                    | :none:`unclaimed`         | :none:`unclaimed`         |                                                                          |
 +-------------------------------------------------------------+---------------------------+---------------------------+--------------------------------------------------------------------------+
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -591,6 +591,7 @@ OpenMP Support
 - Added support 'no_openmp_constructs' assumption clause.
 - Added support for 'self_maps' in map and requirement clause.
 - Added support for 'omp stripe' directive.
+- Added support for threadset clause in task and taskloop directives.
 
 Improvements
 ^^^^^^^^^^^^
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -3691,6 +3691,7 @@ CGOpenMPRuntime::emitTaskInit(CodeGenFunction &CGF, SourceLocation Loc,
     DestructorsFlag = 0x8,
     PriorityFlag = 0x20,
     DetachableFlag = 0x40,
+    PoolFlag = 0x80,
   };
   unsigned Flags = Data.Tied ? TiedFlag : 0;
   bool NeedsCleanup = false;
@@ -3700,6 +3701,11 @@ CGOpenMPRuntime::emitTaskInit(CodeGenFunction &CGF, SourceLocation Loc,
     if (NeedsCleanup)
       Flags = Flags | DestructorsFlag;
   }
+  if (const auto *Clause = D.getSingleClause<OMPThreadsetClause>()) {
+    OpenMPThreadsetKind Kind = Clause->getThreadsetKind();
+    if (Kind == OMPC_THREADSET_omp_pool)
+      Flags = Flags | PoolFlag;
+  }
   if (Data.Priority.getInt())
     Flags = Flags | PriorityFlag;
   if (D.hasClausesOfKind<OMPDetachClause>())
diff --git a/clang/lib/Serialization/ASTReader.cpp b/clang/lib/Serialization/ASTReader.cpp
@@ -11050,6 +11050,9 @@ OMPClause *OMPClauseReader::readClause() {
   case llvm::omp::OMPC_mergeable:
     C = new (Context) OMPMergeableClause();
     break;
+  case llvm::omp::OMPC_threadset:
+    C = new (Context) OMPThreadsetClause();
+    break;
   case llvm::omp::OMPC_read:
     C = new (Context) OMPReadClause();
     break;
diff --git a/clang/test/OpenMP/task_ast_print.cpp b/clang/test/OpenMP/task_ast_print.cpp
@@ -103,11 +103,9 @@ T tmain(T argc, T *argv) {
   a = 2;
 #pragma omp task default(none), private(argc, b) firstprivate(argv) shared(d) if (argc > 0) final(S<T>::TS > 0) priority(argc) affinity(argc, argv[b:argc], arr[:], ([argc][sizeof(T)])argv)
   foo();
-#ifndef OMP60
-#pragma omp taskgroup task_reduction(-: argc)
-#pragma omp task if (C) mergeable priority(C) in_reduction(-: argc)
+#pragma omp taskgroup task_reduction(+: argc)
+#pragma omp task if (C) mergeable priority(C) in_reduction(+: argc)
   foo();
-#endif
   return 0;
 }
 
@@ -123,8 +121,8 @@ T tmain(T argc, T *argv) {
 // CHECK-NEXT: a = 2;
 // CHECK-NEXT: #pragma omp task default(none) private(argc,b) firstprivate(argv) shared(d) if(argc > 0) final(S<T>::TS > 0) priority(argc) affinity(argc,argv[b:argc],arr[:],([argc][sizeof(T)])argv)
 // CHECK-NEXT: foo()
-// CHECK-NEXT: #pragma omp taskgroup task_reduction(-: argc)
-// CHECK-NEXT: #pragma omp task if(C) mergeable priority(C) in_reduction(-: argc)
+// CHECK-NEXT: #pragma omp taskgroup task_reduction(+: argc)
+// CHECK-NEXT: #pragma omp task if(C) mergeable priority(C) in_reduction(+: argc)
 // CHECK-NEXT: foo()
 // CHECK: template<> int tmain<int, 5>(int argc, int *argv) {
 // CHECK-NEXT: int b = argc, c, d, e, f, g;
@@ -138,8 +136,8 @@ T tmain(T argc, T *argv) {
 // CHECK-NEXT: a = 2;
 // CHECK-NEXT: #pragma omp task default(none) private(argc,b) firstprivate(argv) shared(d) if(argc > 0) final(S<int>::TS > 0) priority(argc) affinity(argc,argv[b:argc],arr[:],([argc][sizeof(int)])argv)
 // CHECK-NEXT: foo()
-// CHECK-NEXT: #pragma omp taskgroup task_reduction(-: argc)
-// CHECK-NEXT: #pragma omp task if(5) mergeable priority(5) in_reduction(-: argc)
+// CHECK-NEXT: #pragma omp taskgroup task_reduction(+: argc)
+// CHECK-NEXT: #pragma omp task if(5) mergeable priority(5) in_reduction(+: argc)
 // CHECK-NEXT: foo()
 // CHECK: template<> long tmain<long, 1>(long argc, long *argv) {
 // CHECK-NEXT: long b = argc, c, d, e, f, g;
@@ -153,8 +151,8 @@ T tmain(T argc, T *argv) {
 // CHECK-NEXT: a = 2;
 // CHECK-NEXT: #pragma omp task default(none) private(argc,b) firstprivate(argv) shared(d) if(argc > 0) final(S<long>::TS > 0) priority(argc) affinity(argc,argv[b:argc],arr[:],([argc][sizeof(long)])argv)
 // CHECK-NEXT: foo()
-// CHECK-NEXT: #pragma omp taskgroup task_reduction(-: argc)
-// CHECK-NEXT: #pragma omp task if(1) mergeable priority(1) in_reduction(-: argc)
+// CHECK-NEXT: #pragma omp taskgroup task_reduction(+: argc)
+// CHECK-NEXT: #pragma omp task if(1) mergeable priority(1) in_reduction(+: argc)
 // CHECK-NEXT: foo()
 
 enum Enum {};
diff --git a/clang/test/OpenMP/task_codegen.cpp b/clang/test/OpenMP/task_codegen.cpp
@@ -41,6 +41,9 @@
 // RUN: -emit-llvm -o - -DOMP51 | FileCheck %s \
 // RUN: --implicit-check-not="{{__kmpc|__tgt}}"
 
+// RUN: %clang_cc1 -verify -Wno-vla -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=60 -DOMP60 -fopenmp-enable-irbuilder -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK6
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=60 -DOMP60 -fopenmp-enable-irbuilder -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=60 -DOMP60 -fopenmp-enable-irbuilder -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK6
 
 // expected-no-diagnostics
 #ifndef HEADER
@@ -65,6 +68,7 @@ struct S {
   S(const S &s) : a(s.a) {}
   ~S() {}
 };
+
 int a;
 int main() {
   char b;
@@ -147,6 +151,7 @@ int main() {
 
 
 
+
 // s1 = S();
 
 
@@ -215,6 +220,19 @@ void test_omp_all_memory()
   }
 }
 #endif // OMP51
+
+#ifdef OMP60
+void test_threadset()
+{
+#pragma omp task threadset(omp_team)
+  {
+  }
+#pragma omp task threadset(omp_pool)
+  {
+  }
+}
+#endif // OMP60
+
 #endif
 // CHECK1-LABEL: define {{[^@]+}}@main
 // CHECK1-SAME: () #[[ATTR0:[0-9]+]] {
@@ -10243,3 +10261,18 @@ void test_omp_all_memory()
 // CHECK4-51-NEXT:    call void @__cxx_global_var_init()
 // CHECK4-51-NEXT:    ret void
 //
+// CHECK6-LABEL: define void @_Z14test_threadsetv()
+// CHECK6-NEXT:  entry:
+// CHECK6-NEXT:       [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_23:%.*]], align 1
+// CHECK6-NEXT:       [[AGG_CAPTURED2:%.*]] = alloca [[STRUCT_ANON_25:%.*]], align 1
+// CHECK6-NEXT:       call i32 @__kmpc_global_thread_num(ptr @[[GLOB_PTR:[0-9]+]])
+// CHECK6-NEXT:       [[TMP0:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 %omp_global_thread_num, i32 1, i64 40, i64 1, ptr @.omp_task_entry..[[ENTRY1:[0-9]+]])
+// CHECK6-NEXT:       getelementptr inbounds nuw %struct.kmp_task_t_with_privates{{.*}}, ptr %0, i32 0, i32 0
+// CHECK6-NEXT:       call i32 @__kmpc_global_thread_num(ptr @[[GLOB_PTR:[0-9]+]])
+// CHECK6-NEXT:       call i32 @__kmpc_omp_task(ptr @1, i32 %omp_global_thread_num1, ptr %0)
+// CHECK6-NEXT:       call i32 @__kmpc_global_thread_num(ptr @[[GLOB_PTR2:[0-9]+]])
+// CHECK6-NEXT:       [[TMP3:%.*]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 %omp_global_thread_num3, i32 129, i64 40, i64 1, ptr @.omp_task_entry..[[ENTRY2:[0-9]+]])
+// CHECK6-NEXT:       getelementptr inbounds nuw %struct.kmp_task_t_with_privates{{.*}}, ptr %3, i32 0, i32 0
+// CHECK6-NEXT:       call i32 @__kmpc_global_thread_num(ptr @[[GLOB_PTR2:[0-9]+]])
+// CHECK6-NEXT:       call i32 @__kmpc_omp_task(ptr @1, i32 %omp_global_thread_num4, ptr %3)
+// CHECK6-NEXT:       ret void
diff --git a/clang/test/OpenMP/taskloop_codegen.cpp b/clang/test/OpenMP/taskloop_codegen.cpp
@@ -5,7 +5,12 @@
 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+
 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
+
+// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=60 -DOMP60 -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK6
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=60 -DOMP60 -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=60 -DOMP60 -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK6
 // expected-no-diagnostics
 #ifndef HEADER
 #define HEADER
@@ -241,4 +246,52 @@ void taskloop_with_class() {
   }
 }
 
+#ifdef OMP60
+void test_threadset()
+{
+#pragma omp taskloop threadset(omp_team)
+  for (int i = 0; i < 10; ++i) {
+  }
+#pragma omp taskloop threadset(omp_pool)
+  for (int i = 0; i < 10; ++i) {
+  }
+}
+#endif // OMP60
+// CHECK6-LABEL: define void @_Z14test_threadsetv()
+// CHECK6-NEXT:  entry:
+// CHECK6-NEXT:       [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON_14:%.*]], align 1
+// CHECK6-NEXT:       %[[TMP:.*]] = alloca i32, align 4
+// CHECK6-NEXT:       [[AGG_CAPTURED1:%.*]] = alloca [[STRUCT_ANON_16:%.*]], align 1
+// CHECK6-NEXT:       %[[TMP2:.*]] = alloca i32, align 4
+// CHECK6-NEXT:       %[[TID0:.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB_PTR:[0-9]+]])
+// CHECK6-NEXT:       call void @__kmpc_taskgroup(ptr @1, i32 %[[TID0:.*]])
+// CHECK6-NEXT:       %[[TID1:.*]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 %[[TID0:.*]], i32 1, i64 80, i64 1, ptr @.omp_task_entry..[[ENTRY1:[0-9]+]])
+// CHECK6-NEXT:       %[[TID2:.*]] = getelementptr inbounds nuw %struct.kmp_task_t_with_privates{{.*}}, ptr %[[TID1:.*]], i32 0, i32 0
+// CHECK6-NEXT:       %[[TID3:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID2:.*]], i32 0, i32 5
+// CHECK6-NEXT:       store i64 0, ptr %[[TID3:.*]], align 8
+// CHECK6-NEXT:       %[[TID4:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID2:.*]], i32 0, i32 6
+// CHECK6-NEXT:       store i64 9, ptr %[[TID4:.*]], align 8
+// CHECK6-NEXT:       %[[TID5:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID2:.*]], i32 0, i32 7
+// CHECK6-NEXT:       store i64 1, ptr %[[TID5:.*]], align 8
+// CHECK6-NEXT:       %[[TID6:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID2:.*]], i32 0, i32 9
+// CHECK6-NEXT:       call void @llvm.memset.p0.i64(ptr align 8 %[[TID6:.*]], i8 0, i64 8, i1 false)
+// CHECK6-NEXT:       %[[TID7:.*]] = load i64, ptr %[[TID5:.*]], align 8
+// CHECK6-NEXT:       call void @__kmpc_taskloop(ptr @1, i32 %[[TID0:.*]], ptr %[[TID1:.*]], i32 1, ptr %[[TID3:.*]], ptr %4, i64 %[[TID7:.*]], i32 1, i32 0, i64 0, ptr null)
+// CHECK6-NEXT:       call void @__kmpc_end_taskgroup(ptr @1, i32 %[[TID0:.*]])
+// CHECK6-NEXT:       call void @__kmpc_taskgroup(ptr @1, i32 %[[TID0:.*]])
+// CHECK6-NEXT:       %[[TID8:.*]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 %[[TID0:.*]], i32 129, i64 80, i64 1, ptr @.omp_task_entry..[[ENTRY1:[0-9]+]])
+// CHECK6-NEXT:       %[[TID9:.*]] = getelementptr inbounds nuw %struct.kmp_task_t_with_privates{{.*}}, ptr %[[TID8:.*]], i32 0, i32 0
+// CHECK6-NEXT:       %[[TID10:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID9:.*]], i32 0, i32 5
+// CHECK6-NEXT:       store i64 0, ptr %[[TID10:.*]], align 8
+// CHECK6-NEXT:       %[[TID11:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID9:.*]], i32 0, i32 6
+// CHECK6-NEXT:       store i64 9, ptr %[[TID11:.*]], align 8
+// CHECK6-NEXT:       %[[TID12:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID9:.*]], i32 0, i32 7
+// CHECK6-NEXT:       store i64 1, ptr %[[TID12:.*]], align 8
+// CHECK6-NEXT:       %[[TID13:.*]] = getelementptr inbounds nuw %struct.kmp_task_t{{.*}}, ptr %[[TID9:.*]], i32 0, i32 9
+// CHECK6-NEXT:       call void @llvm.memset.p0.i64(ptr align 8 [[TID13:.*]], i8 0, i64 8, i1 false)
+// CHECK6-NEXT:       %[[TID14:.*]] = load i64, ptr [[TID12:.*]], align 8
+// CHECK6-NEXT:       call void @__kmpc_taskloop(ptr @1, i32 %[[TID0:.*]], ptr %[[TID8:.*]], i32 1, ptr %[[TID10:.*]], ptr %[[TID11:.*]], i64 %[[TID14:.*]], i32 1, i32 0, i64 0, ptr null)
+// CHECK6-NEXT:       call void @__kmpc_end_taskgroup(ptr @1, i32 %[[TID0:.*]])
+// CHECK6-NEXT:       ret void
+
 #endif
