Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/data_sharing.cu
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/data_sharing.cu
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/data_sharing.cu
@@ -553,8 +553,7 @@
     if (GetThreadIdInBlock() == 0) {
       *frame = omptarget_nvptx_simpleMemoryManager.Acquire(buf, size);
     }
-    // FIXME: use __syncthreads instead when the function copy is fixed in LLVM.
-    __SYNCTHREADS();
+    __kmpc_impl_syncthreads();
     return;
   }
   ASSERT0(LT_FUSSY, GetThreadIdInBlock() == GetMasterThreadID(),
@@ -568,8 +567,7 @@
   if (is_shared)
     return;
   if (isSPMDExecutionMode) {
-    // FIXME: use __syncthreads instead when the function copy is fixed in LLVM.
-    __SYNCTHREADS();
+    __kmpc_impl_syncthreads();
     if (GetThreadIdInBlock() == 0) {
       omptarget_nvptx_simpleMemoryManager.Release();
     }
Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
@@ -56,14 +56,6 @@
 #define __ACTIVEMASK() __ballot(1)
 #endif // CUDA_VERSION
 
-#define __SYNCTHREADS_N(n) asm volatile("bar.sync %0;" : : "r"(n) : "memory");
-// Use original __syncthreads if compiled by nvcc or clang >= 9.0.
-#if !defined(__clang__) || __clang_major__ >= 9
-#define __SYNCTHREADS() __syncthreads()
-#else
-#define __SYNCTHREADS() __SYNCTHREADS_N(0)
-#endif
-
 // arguments needed for L0 parallelism only.
 class omptarget_nvptx_SharedArgs {
 public:
Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.cu
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.cu
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.cu
@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "omptarget-nvptx.h"
+#include "target_impl.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // global data tables
@@ -106,7 +107,7 @@
   }
   if (!RequiresOMPRuntime) {
     // Runtime is not required - exit.
-    __SYNCTHREADS();
+    __kmpc_impl_syncthreads();
     return;
   }
 
@@ -125,8 +126,7 @@
     // init team context
     currTeamDescr.InitTeamDescr();
   }
-  // FIXME: use __syncthreads instead when the function copy is fixed in LLVM.
-  __SYNCTHREADS();
+  __kmpc_impl_syncthreads();
 
   omptarget_nvptx_TeamDescr &currTeamDescr = getMyTeamDescriptor();
   omptarget_nvptx_WorkDescr &workDescr = getMyWorkDescriptor();
@@ -168,8 +168,7 @@
   if (!RequiresOMPRuntime)
     return;
 
-  // FIXME: use __syncthreads instead when the function copy is fixed in LLVM.
-  __SYNCTHREADS();
+  __kmpc_impl_syncthreads();
   int threadId = GetThreadIdInBlock();
   if (threadId == 0) {
     // Enqueue omp state object for use by another team.
Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/sync.cu
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/sync.cu
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/sync.cu
@@ -75,8 +75,7 @@
 // parallel region and that all worker threads participate.
 EXTERN void __kmpc_barrier_simple_spmd(kmp_Ident *loc_ref, int32_t tid) {
   PRINT0(LD_SYNC, "call kmpc_barrier_simple_spmd\n");
-  // FIXME: use __syncthreads instead when the function copy is fixed in LLVM.
-  __SYNCTHREADS();
+  __kmpc_impl_syncthreads();
   PRINT0(LD_SYNC, "completed kmpc_barrier_simple_spmd\n");
 }
 
Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h
@@ -63,6 +63,15 @@
 #endif // CUDA_VERSION
 }
 
+INLINE void __kmpc_impl_syncthreads() {
+  // Use original __syncthreads if compiled by nvcc or clang >= 9.0.
+#if !defined(__clang__) || __clang_major__ >= 9
+  __syncthreads();
+#else
+  asm volatile("bar.sync %0;" : : "r"(0) : "memory");
+#endif // __clang__
+}
+
 INLINE void __kmpc_impl_syncwarp(__kmpc_impl_lanemask_t Mask) {
 #if CUDA_VERSION >= 9000
   __syncwarp(Mask);