diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -9967,7 +9967,7 @@
     // Check the error code and execute the host version if required.
     CGF.Builder.restoreIP(OMPBuilder.emitTargetKernel(
         CGF.Builder, Return, RTLoc, DeviceID, NumTeams, NumThreads,
-        OutlinedFnID, KernelArgs));
+        OutlinedFnID, KernelArgs, &CGF.CurFn->getEntryBlock()));
 
     llvm::BasicBlock *OffloadFailedBlock =
         CGF.createBasicBlock("omp_offload.failed");
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -995,10 +995,12 @@
   /// \param NumThreads Number of threads via the 'thread_limit' clause.
   /// \param HostPtr Pointer to the host-side pointer of the target kernel.
   /// \param KernelArgs Array of arguments to the kernel.
+  /// \param EntryBlock Entry basic block of the enclosing function.
   InsertPointTy emitTargetKernel(const LocationDescription &Loc, Value *&Return,
                                  Value *Ident, Value *DeviceID, Value *NumTeams,
                                  Value *NumThreads, Value *HostPtr,
-                                 ArrayRef<Value *> KernelArgs);
+                                 ArrayRef<Value *> KernelArgs,
+                                 llvm::BasicBlock *EntryBlock);
 
   /// Generate a barrier runtime call.
   ///
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -815,12 +815,18 @@
 OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::emitTargetKernel(
     const LocationDescription &Loc, Value *&Return, Value *Ident,
     Value *DeviceID, Value *NumTeams, Value *NumThreads, Value *HostPtr,
-    ArrayRef<Value *> KernelArgs) {
+    ArrayRef<Value *> KernelArgs, llvm::BasicBlock *EntryBlock) {
   if (!updateToLocation(Loc))
     return Loc.IP;
 
-  auto *KernelArgsPtr =
-      Builder.CreateAlloca(OpenMPIRBuilder::KernelArgs, nullptr, "kernel_args");
+  // Insert the alloca at the start of the function entry block.
+  AllocaInst *KernelArgsPtr = nullptr;
+  {
+    IRBuilder<>::InsertPointGuard IPG(Builder);
+    Builder.SetInsertPoint(EntryBlock, EntryBlock->begin());
+    KernelArgsPtr = Builder.CreateAlloca(OpenMPIRBuilder::KernelArgs, nullptr,
+                                         "kernel_args");
+  }
   for (unsigned I = 0, Size = KernelArgs.size(); I != Size; ++I) {
     llvm::Value *Arg =
         Builder.CreateStructGEP(OpenMPIRBuilder::KernelArgs, KernelArgsPtr, I);