diff --git a/openmp/libomptarget/deviceRTLs/common/src/loop.cu b/openmp/libomptarget/deviceRTLs/common/src/loop.cu
--- a/openmp/libomptarget/deviceRTLs/common/src/loop.cu
+++ b/openmp/libomptarget/deviceRTLs/common/src/loop.cu
@@ -14,6 +14,7 @@
 
 #include "common/omptarget.h"
 #include "target_impl.h"
+#include "common/target_atomic.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 ////////////////////////////////////////////////////////////////////////////////
@@ -397,7 +398,7 @@
     unsigned int rank = __kmpc_impl_popc(active & lane_mask_lt);
     uint64_t warp_res;
     if (rank == 0) {
-      warp_res = atomicAdd(
+      warp_res = kmpcAtomicAdd(
           (unsigned long long *)&omptarget_nvptx_threadPrivateContext->Cnt(),
           change);
     }
diff --git a/openmp/libomptarget/deviceRTLs/common/state-queuei.h b/openmp/libomptarget/deviceRTLs/common/state-queuei.h
--- a/openmp/libomptarget/deviceRTLs/common/state-queuei.h
+++ b/openmp/libomptarget/deviceRTLs/common/state-queuei.h
@@ -17,15 +17,16 @@
 //===----------------------------------------------------------------------===//
 
 #include "state-queue.h"
+#include "common/target_atomic.h"
 
 template <typename ElementType, uint32_t SIZE>
 INLINE uint32_t omptarget_nvptx_Queue<ElementType, SIZE>::ENQUEUE_TICKET() {
-  return atomicAdd((unsigned int *)&tail, 1);
+  return kmpcAtomicAdd((unsigned int *)&tail, 1);
 }
 
 template <typename ElementType, uint32_t SIZE>
 INLINE uint32_t omptarget_nvptx_Queue<ElementType, SIZE>::DEQUEUE_TICKET() {
-  return atomicAdd((unsigned int *)&head, 1);
+  return kmpcAtomicAdd((unsigned int *)&head, 1);
 }
 
 template <typename ElementType, uint32_t SIZE>
@@ -37,7 +38,7 @@
 template <typename ElementType, uint32_t SIZE>
 INLINE bool omptarget_nvptx_Queue<ElementType, SIZE>::IsServing(uint32_t slot,
                                                                 uint32_t id) {
-  return atomicAdd((unsigned int *)&ids[slot], 0) == id;
+  return kmpcAtomicAdd((unsigned int *)&ids[slot], 0) == id;
 }
 
 template <typename ElementType, uint32_t SIZE>
@@ -51,8 +52,8 @@
 template <typename ElementType, uint32_t SIZE>
 INLINE ElementType *
 omptarget_nvptx_Queue<ElementType, SIZE>::PopElement(uint32_t slot) {
-  return (ElementType *)atomicAdd((unsigned long long *)&elementQueue[slot],
-                                  (unsigned long long)0);
+  return (ElementType *)kmpcAtomicAdd((unsigned long long *)&elementQueue[slot],
+                                      (unsigned long long)0);
 }
 
 template <typename ElementType, uint32_t SIZE>
diff --git a/openmp/libomptarget/deviceRTLs/common/target_atomic.h b/openmp/libomptarget/deviceRTLs/common/target_atomic.h
new file mode 100644
--- /dev/null
+++ b/openmp/libomptarget/deviceRTLs/common/target_atomic.h
@@ -0,0 +1,22 @@
+//===---- target_atomic.h - OpenMP GPU target atomic functions ---- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Declarations of atomic functions provided by each target
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef OMPTARGET_TARGET_ATOMIC_H
+#define OMPTARGET_TARGET_ATOMIC_H
+
+#include <stdint.h>
+
+uint32_t kmpcAtomicAdd(uint32_t *address, uint32_t val);
+int32_t kmpcAtomicAdd(int32_t *address, int32_t val);
+uint64_t kmpcAtomicAdd(uint64_t *address, uint64_t val);
+
+#endif
diff --git a/openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt b/openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt
--- a/openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt
+++ b/openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt
@@ -59,6 +59,7 @@
       ${devicertl_common_directory}/src/omptarget.cu
       ${devicertl_common_directory}/src/parallel.cu
       src/reduction.cu
+      src/target_atomic.cu
       ${devicertl_common_directory}/src/support.cu
       ${devicertl_common_directory}/src/sync.cu
       ${devicertl_common_directory}/src/task.cu
diff --git a/openmp/libomptarget/deviceRTLs/nvptx/src/libcall.cu b/openmp/libomptarget/deviceRTLs/nvptx/src/libcall.cu
--- a/openmp/libomptarget/deviceRTLs/nvptx/src/libcall.cu
+++ b/openmp/libomptarget/deviceRTLs/nvptx/src/libcall.cu
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "common/omptarget.h"
+#include "common/target_atomic.h"
 
 // Timer precision is 1ns
 #define TIMER_PRECISION ((double)1E-9)
@@ -382,7 +383,7 @@
 EXTERN int omp_test_lock(omp_lock_t *lock) {
   // int atomicCAS(int* address, int compare, int val);
   // (old == compare ? val : old)
-  int ret = atomicAdd(lock, 0);
+  int ret = kmpcAtomicAdd(lock, 0);
 
   PRINT(LD_IO, "call omp_test_lock() return %d\n", ret);
 
diff --git a/openmp/libomptarget/deviceRTLs/nvptx/src/reduction.cu b/openmp/libomptarget/deviceRTLs/nvptx/src/reduction.cu
--- a/openmp/libomptarget/deviceRTLs/nvptx/src/reduction.cu
+++ b/openmp/libomptarget/deviceRTLs/nvptx/src/reduction.cu
@@ -15,6 +15,7 @@
 #include <stdio.h>
 
 #include "common/omptarget.h"
+#include "common/target_atomic.h"
 #include "target_impl.h"
 
 EXTERN
@@ -435,7 +436,7 @@
   bool IsMaster = isMaster(loc, ThreadId);
   while (IsMaster) {
     // Atomic read
-    Bound = atomicAdd((uint32_t *)&IterCnt, 0);
+    Bound = kmpcAtomicAdd((uint32_t *)&IterCnt, 0);
     if (TeamId < Bound + num_of_records)
       break;
   }
@@ -526,7 +527,7 @@
   if (IsMaster && ChunkTeamCount == num_of_records - 1) {
     // Allow SIZE number of teams to proceed writing their
     // intermediate results to the global buffer.
-    atomicAdd((uint32_t *)&IterCnt, num_of_records);
+    kmpcAtomicAdd((uint32_t *)&IterCnt, num_of_records);
   }
 
   return 0;
diff --git a/openmp/libomptarget/deviceRTLs/nvptx/src/target_atomic.cu b/openmp/libomptarget/deviceRTLs/nvptx/src/target_atomic.cu
new file mode 100644
--- /dev/null
+++ b/openmp/libomptarget/deviceRTLs/nvptx/src/target_atomic.cu
@@ -0,0 +1,25 @@
+//===----- target_atomic.cu - NVPTX OpenMP atomic functions ------ CUDA -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Definitions of atomic functions for nvptx
+//===----------------------------------------------------------------------===//
+
+#include "common/target_atomic.h"
+#include "target_impl.h"
+
+uint32_t kmpcAtomicAdd(uint32_t *address, uint32_t val) {
+  return atomicAdd(address, val);
+}
+
+int32_t kmpcAtomicAdd(int32_t *address, int32_t val) {
+  return atomicAdd(address, val);
+}
+
+uint64_t kmpcAtomicAdd(uint64_t *address, uint64_t val) {
+  return atomicAdd(address, val);
+}