diff --git a/paddle/cuda/src/hl_cuda_device.cc b/paddle/cuda/src/hl_cuda_device.cc
index f07538d6ba71334109ef2e2dc572613fbd3cca4e..acd8e2fe6afb4c5642979a6d31f6ec8bbc9b6daa 100644
--- a/paddle/cuda/src/hl_cuda_device.cc
+++ b/paddle/cuda/src/hl_cuda_device.cc
@@ -209,7 +209,18 @@ __thread cudaStream_t default_stream = 0;
 __thread bool g_sync_flag = true;
 bool hl_start_flag = false;
 
-#define gettid() syscall(SYS_gettid)
+inline pid_t gettid() {
+#if defined(__APPLE__) || defined(__OSX__)
+  pid_t tid = syscall(SYS_thread_selfid);
+#else
+  #ifndef __NR_gettid
+  #define __NR_gettid 224
+  #endif
+  pid_t tid = syscall(__NR_gettid);
+#endif
+  CHECK_NE(tid, -1);
+  return tid;    
+}
 
 void hl_init(int device) {
   CHECK(hl_start_flag)
diff --git a/paddle/utils/Thread.h b/paddle/utils/Thread.h
index 3e1d95ab1fcde9d9f1450ec52bf816c517473527..f1352e75d73a098f09864ee2b268d475db7ef8c4 100644
--- a/paddle/utils/Thread.h
+++ b/paddle/utils/Thread.h
@@ -18,7 +18,18 @@ limitations under the License. */
 
 #include <sys/syscall.h>
 #include <unistd.h>
-inline pid_t gettid() { return syscall(SYS_gettid); }
+inline pid_t gettid() {
+#if defined(__APPLE__) || defined(__OSX__)
+  pid_t tid = syscall(SYS_thread_selfid);
+#else
+  #ifndef __NR_gettid
+  #define __NR_gettid 224
+  #endif
+  pid_t tid = syscall(__NR_gettid);
+#endif
+  CHECK_NE(tid, -1);
+  return tid;
+}
 
 #include "Queue.h"
 #include "ThreadLocal.h"
diff --git a/paddle/utils/ThreadLocal.h b/paddle/utils/ThreadLocal.h
index e782868f69a5d0f1a1aeaca1ed2a76001eeb4721..686a1a99a4aa0645cd1a9c636afb9c2c5f50fa58 100644
--- a/paddle/utils/ThreadLocal.h
+++ b/paddle/utils/ThreadLocal.h
@@ -156,7 +156,15 @@ private:
   static void dataDestructor(void* p) { delete (T*)p; }
 
   void updateMap(T* p) {
-    pid_t tid = syscall(SYS_gettid);
+#if defined(__APPLE__) || defined(__OSX__)
+    pid_t tid = syscall(SYS_thread_selfid);
+#else
+    #ifndef __NR_gettid
+    #define __NR_gettid 224
+    #endif
+    pid_t tid = syscall(__NR_gettid);
+#endif
+    CHECK_NE(tid, -1);
     std::lock_guard<std::mutex> guard(mutex_);
     auto ret = threadMap_.insert(std::make_pair(tid, p));
     if (!ret.second) {