add paddle.fluid._cuda_synchronize, test=develop

paddle/fluid/pybind/pybind.cc

@@ -1663,6 +1663,15 @@ All parameter, weight, gradient are variables in Paddle.
   m.def("is_compiled_with_mkldnn", IsCompiledWithMKLDNN);
   m.def("is_compiled_with_brpc", IsCompiledWithBrpc);
   m.def("is_compiled_with_dist", IsCompiledWithDIST);
+  m.def("_cuda_synchronize", [](const platform::XPUPlace &place) {
+    platform::DeviceContextPool::Instance().Get(place)->Wait();
+  });
+  m.def("_cuda_synchronize", [](const platform::CUDAPlace &place) {
+    platform::DeviceContextPool::Instance().Get(place)->Wait();
+  });
+  m.def("_cuda_synchronize", [](const platform::CPUPlace &place) {
+    platform::DeviceContextPool::Instance().Get(place)->Wait();
+  });
 
   m.def("get_float_stats", []() {
     std::vector<paddle::platform::ExportedStatValue<float>> float_stats;

python/paddle/fluid/__init__.py

@@ -90,6 +90,7 @@ from .io import save, load, load_program_state, set_program_state
 from .dygraph.checkpoint import save_dygraph, load_dygraph
 from .dygraph.varbase_patch_methods import monkey_patch_varbase
 from . import generator
+from .core_avx import _cuda_synchronize
 Tensor = LoDTensor
 enable_imperative = enable_dygraph
 disable_imperative = disable_dygraph
@@ -133,7 +134,8 @@ __all__ = framework.__all__ + executor.__all__ + \
         'install_check',
         'save',
         'load',
-        'VarBase'
+        'VarBase',
+        '_cuda_synchronize'
     ]