diff --git a/cmake/flags.cmake b/cmake/flags.cmake
index bcd734b7e4dbcc70211457862425934a38f786be..cc59309ee7efabcff167a0a80a6ad370f159e99b 100644
--- a/cmake/flags.cmake
+++ b/cmake/flags.cmake
@@ -71,6 +71,11 @@ foreach(flag ${COMMON_FLAGS})
     safe_set_cxxflag(CMAKE_CXX_FLAGS ${flag})
 endforeach()
 
+# On Mac OS X build fat binaries with x86_64 architectures by default.
+if (APPLE)
+    set (CMAKE_OSX_ARCHITECTURES "x86_64" CACHE STRING "Build architectures for OSX" FORCE)
+endif ()
+
 # Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
 # So, don't set these flags here.
 
diff --git a/paddle/cuda/src/hl_cuda_device.cc b/paddle/cuda/src/hl_cuda_device.cc
index acd8e2fe6afb4c5642979a6d31f6ec8bbc9b6daa..f4c07367b485b8c8f159a1a6f8e16295c7dd6fb6 100644
--- a/paddle/cuda/src/hl_cuda_device.cc
+++ b/paddle/cuda/src/hl_cuda_device.cc
@@ -211,7 +211,11 @@ bool hl_start_flag = false;
 
 inline pid_t gettid() {
 #if defined(__APPLE__) || defined(__OSX__)
-  pid_t tid = syscall(SYS_thread_selfid);
+  // syscall is deprecated: first deprecated in macOS 10.12.
+  // syscall is unsupported;
+  // syscall pid_t tid = syscall(SYS_thread_selfid);
+  uint64_t tid;
+  pthread_threadid_np(NULL, &tid);
 #else
   #ifndef __NR_gettid
   #define __NR_gettid 224
diff --git a/paddle/pserver/SocketChannel.cpp b/paddle/pserver/SocketChannel.cpp
index b9d542a296ddddcd5e00d345b71fa976f0d72388..20295d7cdc22b5dba6380a0792eafef9feec257a 100644
--- a/paddle/pserver/SocketChannel.cpp
+++ b/paddle/pserver/SocketChannel.cpp
@@ -157,7 +157,8 @@ void SocketChannel::writeMessage(const std::vector<struct iovec>& userIovs) {
   std::vector<iovec> iovs;
   iovs.reserve(userIovs.size() + 2);
   iovs.push_back({&header, sizeof(header)});
-  iovs.push_back({&iovLengths[0], sizeof(iovLengths[0]) * header.numIovs});
+  iovs.push_back({&iovLengths[0], static_cast<size_t>(
+      sizeof(iovLengths[0]) * header.numIovs)});
   iovs.insert(iovs.end(), userIovs.begin(), userIovs.end());
 
   header.totalLength = 0;
diff --git a/paddle/setup.py.in b/paddle/setup.py.in
index 02ea9067431c62263d9a3a21d112c9ea64bacbe4..3341dd6f95969fcd8df5b6049b0b8d2d5905a43f 100644
--- a/paddle/setup.py.in
+++ b/paddle/setup.py.in
@@ -18,6 +18,7 @@ from setuptools import setup, Extension
 import numpy as np
 import api.paddle_ld_flags
 import platform
+import os
 
 system = platform.system().lower()
 
@@ -45,6 +46,7 @@ except:
 if is_lin == True:
     extra_links = ["-Xlinker", '-start-group'] + extra_links + ["-Xlinker", "-end-group"]
 elif is_osx == True:
+    os.environ["ARCHFLAGS"] = "-arch x86_64"
     extra_links = ["-Wl,-all_load"] + extra_links
 
 include_dirs = [np.get_include(), "../"]    # include numpy and paddle.
diff --git a/paddle/utils/ThreadLocal.h b/paddle/utils/ThreadLocal.h
index 686a1a99a4aa0645cd1a9c636afb9c2c5f50fa58..b91e4ad5472cab4f48f1eb59304aa7c0cf3f621f 100644
--- a/paddle/utils/ThreadLocal.h
+++ b/paddle/utils/ThreadLocal.h
@@ -22,6 +22,7 @@ limitations under the License. */
 #include <map>
 #include <mutex>
 #include <random>
+#include "Util.h"
 #include "Logging.h"
 
 namespace paddle {
@@ -156,14 +157,7 @@ private:
   static void dataDestructor(void* p) { delete (T*)p; }
 
   void updateMap(T* p) {
-#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
+    pid_t tid = getTID();
     CHECK_NE(tid, -1);
     std::lock_guard<std::mutex> guard(mutex_);
     auto ret = threadMap_.insert(std::make_pair(tid, p));
diff --git a/paddle/utils/Util.cpp b/paddle/utils/Util.cpp
index c3c76f907d40e0ccebddbe73f183af326dac8b8c..45251213d2d7930947f39d4730245ca8f7dfddc8 100644
--- a/paddle/utils/Util.cpp
+++ b/paddle/utils/Util.cpp
@@ -95,7 +95,11 @@ namespace paddle {
 
 pid_t getTID() {
   #if defined(__APPLE__) || defined(__OSX__)
-      pid_t tid = syscall(SYS_thread_selfid);
+      // syscall is deprecated: first deprecated in macOS 10.12.
+      // syscall is unsupported;
+      // syscall pid_t tid = syscall(SYS_thread_selfid);
+      uint64_t tid;
+      pthread_threadid_np(NULL, &tid);
   #else
       #ifndef __NR_gettid
       #define __NR_gettid 224
diff --git a/paddle/utils/arch/osx/Locks.cpp b/paddle/utils/arch/osx/Locks.cpp
index 47e44e9d7c114cc2a8b038c816b73511c048d82e..44bab7198d5f496786b05c909f263b17f5849234 100644
--- a/paddle/utils/arch/osx/Locks.cpp
+++ b/paddle/utils/arch/osx/Locks.cpp
@@ -16,6 +16,11 @@ limitations under the License. */
 #include "paddle/utils/Logging.h"
 #include <dispatch/dispatch.h>
 #include <libkern/OSAtomic.h>
+
+#if MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_12
+#include <os/lock.h>
+#endif
+
 namespace paddle {
 
 class SemaphorePrivate {
@@ -50,21 +55,32 @@ void Semaphore::post() {
 
 class SpinLockPrivate {
 public:
+#if MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_12
+  os_unfair_lock lock_;
+#else
   SpinLockPrivate(): lock_(OS_SPINLOCK_INIT) {}
-
   OSSpinLock lock_;
-  char padding_[64 - sizeof(OSSpinLock)];  // Padding to cache line size
+#endif
+  char padding_[64 - sizeof(lock_)];  // Padding to cache line size
 };
 
 SpinLock::SpinLock(): m(new SpinLockPrivate()) {}
 SpinLock::~SpinLock() { delete m; }
 
 void SpinLock::lock() {
+#if MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_12
+  os_unfair_lock_lock(&m->lock_);
+#else
   OSSpinLockLock(&m->lock_);
+#endif
 }
 
 void SpinLock::unlock() {
+#if MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_12
+  os_unfair_lock_unlock(&m->lock_);
+#else
   OSSpinLockUnlock(&m->lock_);
+#endif
 }