Add synchronize method to all devices.
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h b/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h
index d9a3d32..7864a2a 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceDefault.h
@@ -95,6 +95,10 @@
     return firstLevelCacheSize();
 #endif
   }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void synchronize() const {
+    // Nothing.  Default device operations are synchronous.
+  }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE int majorDeviceVersion() const {
 #if !defined(EIGEN_GPU_COMPILE_PHASE)
diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceThreadPool.h b/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceThreadPool.h
index a3adf61..6accc66 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceThreadPool.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorDeviceThreadPool.h
@@ -147,6 +147,10 @@
     // The l3 cache size is shared between all the cores.
     return l3CacheSize() / num_threads_;
   }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void synchronize() const {
+    // Nothing.  Threadpool device operations are synchronous.
+  }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE int majorDeviceVersion() const {
     // Should return an enum that encodes the ISA supported by the CPU