2 files changed, 16 insertions, 9 deletions
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 478fda7d913..3730b9af12f 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2839,12 +2839,12 @@ void CGOpenMPRuntime::emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
         RT.createRuntimeFunction(OMPRTL__kmpc_serialized_parallel), Args);
 
     // OutlinedFn(&GTid, &zero, CapturedStruct);
-    Address ThreadIDAddr = RT.emitThreadIDAddress(CGF, Loc);
     Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
                                                         /*Name*/ ".zero.addr");
     CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
     llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
-    OutlinedFnArgs.push_back(ThreadIDAddr.getPointer());
+    // ThreadId for serialized parallels is 0.
+    OutlinedFnArgs.push_back(ZeroAddr.getPointer());
     OutlinedFnArgs.push_back(ZeroAddr.getPointer());
     OutlinedFnArgs.append(CapturedVars.begin(), CapturedVars.end());
     RT.emitOutlinedFunctionCall(CGF, Loc, OutlinedFn, OutlinedFnArgs);
diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
index 5aaebff8f76..036b5371fe0 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -1784,8 +1784,9 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
                                            /*DestWidth=*/32, /*Signed=*/1),
                                        ".zero.addr");
   CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
-  Address ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
-  auto &&CodeGen = [this, Fn, CapturedVars, Loc, ZeroAddr, ThreadIDAddr](
+  // ThreadId for serialized parallels is 0.
+  Address ThreadIDAddr = ZeroAddr;
+  auto &&CodeGen = [this, Fn, CapturedVars, Loc, ZeroAddr, &ThreadIDAddr](
                        CodeGenFunction &CGF, PrePostActionTy &Action) {
     Action.Enter(CGF);
 
@@ -1883,8 +1884,9 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
     Work.emplace_back(WFn);
   };
 
-  auto &&LNParallelGen = [this, Loc, &SeqGen, &L0ParallelGen, &CodeGen](
-                             CodeGenFunction &CGF, PrePostActionTy &Action) {
+  auto &&LNParallelGen = [this, Loc, &SeqGen, &L0ParallelGen, &CodeGen,
+                          &ThreadIDAddr](CodeGenFunction &CGF,
+                                         PrePostActionTy &Action) {
     RegionCodeGenTy RCG(CodeGen);
     if (IsInParallelRegion) {
       SeqGen(CGF, Action);
@@ -1936,6 +1938,8 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
       // There is no need to emit line number for unconditional branch.
       (void)ApplyDebugLocation::CreateEmpty(CGF);
       CGF.EmitBlock(ElseBlock);
+      // In the worker need to use the real thread id.
+      ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
       RCG(CGF);
       // There is no need to emit line number for unconditional branch.
       (void)ApplyDebugLocation::CreateEmpty(CGF);
@@ -1965,10 +1969,11 @@ void CGOpenMPRuntimeNVPTX::emitSPMDParallelCall(
                                            /*DestWidth=*/32, /*Signed=*/1),
                                        ".zero.addr");
   CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
-  Address ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
+  // ThreadId for serialized parallels is 0.
+  Address ThreadIDAddr = ZeroAddr;
   auto &&CodeGen = [this, OutlinedFn, CapturedVars, Loc, ZeroAddr,
-                    ThreadIDAddr](CodeGenFunction &CGF,
-                                  PrePostActionTy &Action) {
+                    &ThreadIDAddr](CodeGenFunction &CGF,
+                                   PrePostActionTy &Action) {
     Action.Enter(CGF);
 
     llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
@@ -1995,6 +2000,8 @@ void CGOpenMPRuntimeNVPTX::emitSPMDParallelCall(
   };
 
   if (IsInTargetMasterThreadRegion) {
+    // In the worker need to use the real thread id.
+    ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
     RegionCodeGenTy RCG(CodeGen);
     RCG(CGF);
   } else {