[OPENMP] Fix emission of the __kmpc_global_thread_num.

Fixed emission of the __kmpc_global_thread_num() so that it is not
messed up with alloca instructions anymore. Plus, fixes emission of the
__kmpc_global_thread_num() functions in the target outlined regions so
that they are not called before runtime is initialized.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@343856 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGOpenMPRuntime.cpp b/lib/CodeGen/CGOpenMPRuntime.cpp
index d421729faa2ec5bda4ee157c295f2cf62e804613..c00f66dd0301185c579f6b10734e097d5f873228 100644 (file)
--- a/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -1485,6 +1485,31 @@ Address CGOpenMPRuntime::getOrCreateDefaultLocation(unsigned Flags) {
   return Address(Entry, Align);
 }
 
+void CGOpenMPRuntime::setLocThreadIdInsertPt(CodeGenFunction &CGF,
+                                             bool AtCurrentPoint) {
+  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
+  assert(!Elem.second.ServiceInsertPt && "Insert point is set already.");
+
+  llvm::Value *Undef = llvm::UndefValue::get(CGF.Int32Ty);
+  if (AtCurrentPoint) {
+    Elem.second.ServiceInsertPt = new llvm::BitCastInst(
+        Undef, CGF.Int32Ty, "svcpt", CGF.Builder.GetInsertBlock());
+  } else {
+    Elem.second.ServiceInsertPt =
+        new llvm::BitCastInst(Undef, CGF.Int32Ty, "svcpt");
+    Elem.second.ServiceInsertPt->insertAfter(CGF.AllocaInsertPt);
+  }
+}
+
+void CGOpenMPRuntime::clearLocThreadIdInsertPt(CodeGenFunction &CGF) {
+  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
+  if (Elem.second.ServiceInsertPt) {
+    llvm::Instruction *Ptr = Elem.second.ServiceInsertPt;
+    Elem.second.ServiceInsertPt = nullptr;
+    Ptr->eraseFromParent();
+  }
+}
+
 llvm::Value *CGOpenMPRuntime::emitUpdateLocation(CodeGenFunction &CGF,
                                                  SourceLocation Loc,
                                                  unsigned Flags) {
@@ -1511,8 +1536,10 @@ llvm::Value *CGOpenMPRuntime::emitUpdateLocation(CodeGenFunction &CGF,
     Elem.second.DebugLoc = AI.getPointer();
     LocValue = AI;
 
+    if (!Elem.second.ServiceInsertPt)
+      setLocThreadIdInsertPt(CGF);
     CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
-    CGF.Builder.SetInsertPoint(CGF.AllocaInsertPt);
+    CGF.Builder.SetInsertPoint(Elem.second.ServiceInsertPt);
     CGF.Builder.CreateMemCpy(LocValue, getOrCreateDefaultLocation(Flags),
                              CGF.getTypeSize(IdentQTy));
   }
@@ -1582,21 +1609,25 @@ llvm::Value *CGOpenMPRuntime::getThreadID(CodeGenFunction &CGF,
   // kmpc_global_thread_num(ident_t *loc).
   // Generate thread id value and cache this value for use across the
   // function.
+  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
+  if (!Elem.second.ServiceInsertPt)
+    setLocThreadIdInsertPt(CGF);
   CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
-  CGF.Builder.SetInsertPoint(CGF.AllocaInsertPt);
+  CGF.Builder.SetInsertPoint(Elem.second.ServiceInsertPt);
   llvm::CallInst *Call = CGF.Builder.CreateCall(
       createRuntimeFunction(OMPRTL__kmpc_global_thread_num),
       emitUpdateLocation(CGF, Loc));
   Call->setCallingConv(CGF.getRuntimeCC());
-  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
   Elem.second.ThreadID = Call;
   return Call;
 }
 
 void CGOpenMPRuntime::functionFinished(CodeGenFunction &CGF) {
   assert(CGF.CurFn && "No function in current CodeGenFunction.");
-  if (OpenMPLocThreadIDMap.count(CGF.CurFn))
+  if (OpenMPLocThreadIDMap.count(CGF.CurFn)) {
+    clearLocThreadIdInsertPt(CGF);
     OpenMPLocThreadIDMap.erase(CGF.CurFn);
+  }
   if (FunctionUDRMap.count(CGF.CurFn) > 0) {
     for(auto *D : FunctionUDRMap[CGF.CurFn])
       UDRMap.erase(D);

diff --git a/lib/CodeGen/CGOpenMPRuntime.h b/lib/CodeGen/CGOpenMPRuntime.h
index 35f75a9ec0861654d82e561321e719733b630a80..3fb1a1538fa6bd690b20e493011ff06dff87397a 100644 (file)
--- a/lib/CodeGen/CGOpenMPRuntime.h
+++ b/lib/CodeGen/CGOpenMPRuntime.h
@@ -278,6 +278,10 @@ protected:
   /// stored.
   virtual Address emitThreadIDAddress(CodeGenFunction &CGF, SourceLocation Loc);
 
+  void setLocThreadIdInsertPt(CodeGenFunction &CGF,
+                              bool AtCurrentPoint = false);
+  void clearLocThreadIdInsertPt(CodeGenFunction &CGF);
+
 private:
   /// Default const ident_t object used for initialization of all other
   /// ident_t objects.
@@ -300,6 +304,8 @@ private:
   struct DebugLocThreadIdTy {
     llvm::Value *DebugLoc;
     llvm::Value *ThreadID;
+    /// Insert point for the service instructions.
+    llvm::AssertingVH<llvm::Instruction> ServiceInsertPt = nullptr;
   };
   /// Map of local debug location, ThreadId and functions.
   typedef llvm::DenseMap<llvm::Function *, DebugLocThreadIdTy>

diff --git a/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp b/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index a923232e8173d43931a543789a8ee7ffc0016810..f0f0a735f8a79fc1aa997d835f38050bae7ececd 100644 (file)
--- a/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -1197,8 +1197,11 @@ void CGOpenMPRuntimeNVPTX::emitSPMDKernel(const OMPExecutableDirective &D,
         : RT(RT), EST(EST), D(D) {}
     void Enter(CodeGenFunction &CGF) override {
       RT.emitSPMDEntryHeader(CGF, EST, D);
+      // Skip target region initialization.
+      RT.setLocThreadIdInsertPt(CGF, /*AtCurrentPoint=*/true);
     }
     void Exit(CodeGenFunction &CGF) override {
+      RT.clearLocThreadIdInsertPt(CGF);
       RT.emitSPMDEntryFooter(CGF, EST);
     }
   } Action(*this, EST, D);

diff --git a/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
index d75a10210905b2aec54c67fa2ba4e740ce35c6cf..946447532539333000918806f2fd448a99bc6414 100644 (file)
--- a/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
+++ b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp
@@ -22,8 +22,8 @@ int main(int argc, char **argv) {
 // CHECK: @__omp_offloading_{{.*}}_main_l16_exec_mode = weak constant i8 0
 
 // CHECK: define weak void @__omp_offloading_{{.*}}_main_l16(i{{64|32}} %{{[^,].*}}, i32* dereferenceable{{[^,]*}}, i{{64|32}} %{{[^,)]*}})
-// CHECK: [[TID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @
 // CHECK: call void @__kmpc_spmd_kernel_init(
+// CHECK: [[TID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @
 // CHECK: call void @__kmpc_for_static_init_4(
 
 // CHECK: call void [[PARALLEL:@.+]](i32* %{{.*}}, i32* %{{.+}}, i{{64|32}} %{{.+}}, i{{64|32}} %{{.*}}, i{{64|32}} %{{.*}}, i32* %{{.*}})

diff --git a/test/OpenMP/parallel_if_codegen.cpp b/test/OpenMP/parallel_if_codegen.cpp
index 49e4bd402ce8709be785d20dacbb65eac36aa801..ec9fc01561bd212895f96945b8d594fc22dd02a3 100644 (file)
--- a/test/OpenMP/parallel_if_codegen.cpp
+++ b/test/OpenMP/parallel_if_codegen.cpp
@@ -55,9 +55,9 @@ int tmain(T Arg) {
 
 // CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main()
 int main() {
-// CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num(
 // CHECK: store i32 0, i32* [[ZERO_ADDR2:%.+]],
 // CHECK: store i32 0, i32* [[ZERO_ADDR1:%.+]],
+// CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num(
 // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN4:@.+]] to void
 #pragma omp parallel if (true)
   fn4();
@@ -96,9 +96,9 @@ int main() {
 // CHECK: ret void
 
 // CHECK-LABEL: define {{.+}} @{{.+}}tmain
-// CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num(
 // CHECK: store i32 0, i32* [[ZERO_ADDR2:%.+]],
 // CHECK: store i32 0, i32* [[ZERO_ADDR1:%.+]],
+// CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num(
 // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN1:@.+]] to void
 // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
 // CHECK: call void [[CAP_FN2:@.+]](i32* [[ZERO_ADDR1]], i32* [[ZERO_ADDR1]])

diff --git a/test/OpenMP/single_codegen.cpp b/test/OpenMP/single_codegen.cpp
index a5426bf483e281dc2fbdd7ab4a80c1a163dce0f3..df5b2ca0e67aaabe7d35ec2565efeb6207d613b2 100644 (file)
--- a/test/OpenMP/single_codegen.cpp
+++ b/test/OpenMP/single_codegen.cpp
@@ -74,9 +74,12 @@ struct SST {
 // CHECK-LABEL: @main
 // TERM_DEBUG-LABEL: @main
 int main() {
+  // CHECK:     alloca i32
   // CHECK-DAG: [[A_ADDR:%.+]] = alloca i8
   // CHECK-DAG: [[A2_ADDR:%.+]] = alloca [2 x i8]
   // CHECK-DAG: [[C_ADDR:%.+]] = alloca [[TEST_CLASS_TY]]
+  // CHECK-DAG: [[DID_IT:%.+]] = alloca i32,
+  // CHECK-DAG: [[COPY_LIST:%.+]] = alloca [5 x i8*],
   char a;
   char a2[2];
   TestClass &c = tc;
@@ -84,9 +87,6 @@ int main() {
   SS ss(c.a);
 
 // CHECK:       [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:@.+]])
-// CHECK-DAG:   [[DID_IT:%.+]] = alloca i32,
-// CHECK-DAG:   [[COPY_LIST:%.+]] = alloca [5 x i8*],
-
 // CHECK:       [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
 // CHECK-NEXT:  [[IS_SINGLE:%.+]] = icmp ne i32 [[RES]], 0
 // CHECK-NEXT:  br i1 [[IS_SINGLE]], label {{%?}}[[THEN:.+]], label {{%?}}[[EXIT:.+]]

diff --git a/test/OpenMP/single_firstprivate_codegen.cpp b/test/OpenMP/single_firstprivate_codegen.cpp
index a59e1a775a046f121d8694093b4255ad5713111c..3b738253904eb5501897363395733be742a0304e 100644 (file)
--- a/test/OpenMP/single_firstprivate_codegen.cpp
+++ b/test/OpenMP/single_firstprivate_codegen.cpp
@@ -178,13 +178,13 @@ int main() {
 
 // CHECK: define {{.*}}i{{[0-9]+}} @main()
 // CHECK: alloca i{{[0-9]+}},
-// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
 // CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 // CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}],
 // CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]],
 // CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]],
 // CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}},
 
+// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
 // CHECK: call i32 @__kmpc_single(
 // firstprivate t_var(t_var)
 // CHECK: [[T_VAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR]],

diff --git a/test/OpenMP/taskgroup_task_reduction_codegen.cpp b/test/OpenMP/taskgroup_task_reduction_codegen.cpp
index 90b03c653ef027c014c221c14e285f72042d2e28..02a5ba664e5e9ee9be6f914b30bf9f60a63fdd2c 100644 (file)
--- a/test/OpenMP/taskgroup_task_reduction_codegen.cpp
+++ b/test/OpenMP/taskgroup_task_reduction_codegen.cpp
@@ -43,12 +43,12 @@ int main(int argc, char **argv) {
 // CHECK:       [[A:%.+]] = alloca i32,
 // CHECK:       [[B:%.+]] = alloca float,
 // CHECK:       [[C:%.+]] = alloca [5 x %struct.S],
-// CHECK:       [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
 // CHECK:       [[RD_IN1:%.+]] = alloca [3 x [[T1:%[^,]+]]],
 // CHECK:       [[TD1:%.+]] = alloca i8*,
 // CHECK:       [[RD_IN2:%.+]] = alloca [2 x [[T2:%[^,]+]]],
 // CHECK:       [[TD2:%.+]] = alloca i8*,
 
+// CHECK:       [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
 // CHECK:       [[VLA:%.+]] = alloca i16, i64 [[VLA_SIZE:%[^,]+]],
 
 // CHECK:       call void @__kmpc_taskgroup(%struct.ident_t* {{[^,]+}}, i32 [[GTID]])

diff --git a/test/OpenMP/taskloop_reduction_codegen.cpp b/test/OpenMP/taskloop_reduction_codegen.cpp
index 65bc50a4d25535c5b9200474a5d52951c27b086b..3a3434a4540946c3959d3363c1492ab77b8082d3 100644 (file)
--- a/test/OpenMP/taskloop_reduction_codegen.cpp
+++ b/test/OpenMP/taskloop_reduction_codegen.cpp
@@ -52,11 +52,11 @@ sum = 0.0;
 // CHECK:    [[C:%.*]] = alloca [100 x %struct.S],
 // CHECK:    [[D:%.*]] = alloca float*,
 // CHECK:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]],
-// CHECK:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
 // CHECK:    [[DOTRD_INPUT_:%.*]] = alloca [4 x %struct.kmp_task_red_input_t],
 // CHECK:    alloca i32,
 // CHECK:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32,
 // CHECK:    [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32,
+// CHECK:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
 // CHECK:    store i32 0, i32* [[RETVAL]],
 // CHECK:    store i32 [[ARGC:%.*]], i32* [[ARGC_ADDR]],
 // CHECK:    store i8** [[ARGV:%.*]], i8*** [[ARGV_ADDR]],

diff --git a/test/OpenMP/taskloop_simd_reduction_codegen.cpp b/test/OpenMP/taskloop_simd_reduction_codegen.cpp
index 6c34e34ca584079793ead0a80c1ea1a25af6a1dd..d2c4b8b311378c84ee8a4a3fea54c9d7b3a2d2d3 100644 (file)
--- a/test/OpenMP/taskloop_simd_reduction_codegen.cpp
+++ b/test/OpenMP/taskloop_simd_reduction_codegen.cpp
@@ -52,11 +52,11 @@ sum = 0.0;
 // CHECK:    [[C:%.*]] = alloca [100 x %struct.S],
 // CHECK:    [[D:%.*]] = alloca float*,
 // CHECK:    [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]],
-// CHECK:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
 // CHECK:    [[DOTRD_INPUT_:%.*]] = alloca [4 x %struct.kmp_task_red_input_t],
 // CHECK:    alloca i32,
 // CHECK:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32,
 // CHECK:    [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32,
+// CHECK:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
 // CHECK:    store i32 0, i32* [[RETVAL]],
 // CHECK:    store i32 [[ARGC:%.*]], i32* [[ARGC_ADDR]],
 // CHECK:    store i8** [[ARGV:%.*]], i8*** [[ARGV_ADDR]],