2 files changed, 26 insertions, 19 deletions
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 42d18b4098e..7ff36c79f6f 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -3075,14 +3075,18 @@ void CGOpenMPRuntime::emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
     CGF.EmitRuntimeCall(
         RT.createRuntimeFunction(OMPRTL__kmpc_serialized_parallel), Args);
 
-    // OutlinedFn(&GTid, &zero, CapturedStruct);
+    // OutlinedFn(&zero, &zero_bound, CapturedStruct);
     Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
-                                                        /*Name*/ ".zero.addr");
+                                                        /*Name=*/".zero.addr");
     CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
+    Address ZeroAddrBound =
+        CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
+                                         /*Name=*/".bound.zero.addr");
+    CGF.InitTempAlloca(ZeroAddrBound, CGF.Builder.getInt32(/*C*/ 0));
     llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
     // ThreadId for serialized parallels is 0.
     OutlinedFnArgs.push_back(ZeroAddr.getPointer());
-    OutlinedFnArgs.push_back(ZeroAddr.getPointer());
+    OutlinedFnArgs.push_back(ZeroAddrBound.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 63093e73257..291052109e8 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -2459,9 +2459,8 @@ void CGOpenMPRuntimeNVPTX::emitTeamsCall(CodeGenFunction &CGF,
   if (!CGF.HaveInsertPoint())
     return;
 
-  Address ZeroAddr = CGF.CreateMemTemp(
-      CGF.getContext().getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1),
-      /*Name*/ ".zero.addr");
+  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(emitThreadIDAddress(CGF, Loc).getPointer());
@@ -2490,16 +2489,19 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
   // Force inline this outlined function at its call site.
   Fn->setLinkage(llvm::GlobalValue::InternalLinkage);
 
-  Address ZeroAddr = CGF.CreateMemTemp(CGF.getContext().getIntTypeForBitwidth(
-                                           /*DestWidth=*/32, /*Signed=*/1),
-                                       ".zero.addr");
+  Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
+                                                      /*Name=*/".zero.addr");
   CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
   // ThreadId for serialized parallels is 0.
   Address ThreadIDAddr = ZeroAddr;
-  auto &&CodeGen = [this, Fn, CapturedVars, Loc, ZeroAddr, &ThreadIDAddr](
+  auto &&CodeGen = [this, Fn, CapturedVars, Loc, &ThreadIDAddr](
                        CodeGenFunction &CGF, PrePostActionTy &Action) {
     Action.Enter(CGF);
 
+    Address ZeroAddr =
+        CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
+                                         /*Name=*/".bound.zero.addr");
+    CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
     llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
     OutlinedFnArgs.push_back(ThreadIDAddr.getPointer());
     OutlinedFnArgs.push_back(ZeroAddr.getPointer());
@@ -2656,17 +2658,19 @@ void CGOpenMPRuntimeNVPTX::emitSPMDParallelCall(
   //
   llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
 
-  Address ZeroAddr = CGF.CreateMemTemp(CGF.getContext().getIntTypeForBitwidth(
-                                           /*DestWidth=*/32, /*Signed=*/1),
-                                       ".zero.addr");
+  Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
+                                                      /*Name=*/".zero.addr");
   CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
   // ThreadId for serialized parallels is 0.
   Address ThreadIDAddr = ZeroAddr;
-  auto &&CodeGen = [this, OutlinedFn, CapturedVars, Loc, ZeroAddr,
-                    &ThreadIDAddr](CodeGenFunction &CGF,
-                                   PrePostActionTy &Action) {
+  auto &&CodeGen = [this, OutlinedFn, CapturedVars, Loc, &ThreadIDAddr](
+                       CodeGenFunction &CGF, PrePostActionTy &Action) {
     Action.Enter(CGF);
 
+    Address ZeroAddr =
+        CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
+                                         /*Name=*/".bound.zero.addr");
+    CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
     llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
     OutlinedFnArgs.push_back(ThreadIDAddr.getPointer());
     OutlinedFnArgs.push_back(ZeroAddr.getPointer());
@@ -4567,9 +4571,8 @@ llvm::Function *CGOpenMPRuntimeNVPTX::createParallelDataSharingWrapper(
   const auto *RD = CS.getCapturedRecordDecl();
   auto CurField = RD->field_begin();
 
-  Address ZeroAddr = CGF.CreateMemTemp(
-      CGF.getContext().getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1),
-      /*Name*/ ".zero.addr");
+  Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
+                                                      /*Name=*/".zero.addr");
   CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
   // Get the array of arguments.
   SmallVector<llvm::Value *, 8> Args;