disable the openblas multi-thread on windows since no support

adjust the python script

paddle/fluid/platform/cpu_helper.cc

@@ -29,6 +29,12 @@ namespace platform {
 
 void SetNumThreads(int num_threads) {
 #ifdef PADDLE_USE_OPENBLAS
+  // windows has no support for openblas multi-thread
+#ifdef _WIN32
+  if (num_threads > 1) {
+    num_threads = 1;
+  }
+#endif
   int real_num_threads = num_threads > 1 ? num_threads : 1;
   openblas_set_num_threads(real_num_threads);
 #elif defined(PADDLE_WITH_MKLML)

paddle/fluid/platform/init.cc

@@ -113,13 +113,6 @@ void InitDevices(bool init_p2p, const std::vector<int> devices) {
   places.emplace_back(platform::CPUPlace());
   platform::DeviceContextPool::Init(places);
 
-// windows has no support for openblas multi-thread
-#ifdef _WIN32
-  if (FLAGS_paddle_num_threads > 1) {
-    FLAGS_paddle_num_threads = 1;
-  }
-#endif
-
 #ifndef PADDLE_WITH_MKLDNN
   platform::SetNumThreads(FLAGS_paddle_num_threads);
 #endif

python/paddle/fluid/__init__.py

@@ -47,7 +47,8 @@ from . import profiler
 from . import unique_name
 from . import recordio_writer
 from . import parallel_executor
-from .parallel_executor import *
+if os.name != 'nt':
+    from .parallel_executor import *
 from paddle.fluid.layers.math_op_patch import monkey_patch_variable
 
 Tensor = LoDTensor

python/paddle/fluid/contrib/inferencer.py

@@ -15,15 +15,13 @@
 from __future__ import print_function
 
 import contextlib
-import os
 
 from .. import core
 
 from .. import executor
 from .. import framework
 from .. import io
-if os.name != 'nt':
-    from .. import parallel_executor
+from .. import parallel_executor
 from .. import unique_name
 from .trainer import check_and_get_place
 

python/paddle/fluid/contrib/trainer.py

@@ -28,8 +28,7 @@ from .. import framework
 from .. import io
 # optimizer is same as the parameter of Trainer.__init__. Rename it to opt_module
 from .. import optimizer as opt_module
-if os.name != 'nt':
-    from .. import parallel_executor
+from .. import parallel_executor
 from ..transpiler import distribute_transpiler
 
 __all__ = [

python/paddle/fluid/parallel_executor.py

@@ -25,11 +25,12 @@ import os
 
 __all__ = ['ParallelExecutor', 'ExecutionStrategy', 'BuildStrategy']
 
-ExecutionStrategy = core.ParallelExecutor.ExecutionStrategy
-BuildStrategy = core.ParallelExecutor.BuildStrategy
+if os.name != 'nt':
+    ExecutionStrategy = core.ParallelExecutor.ExecutionStrategy
+    BuildStrategy = core.ParallelExecutor.BuildStrategy
 
 
-class ParallelExecutor(object):
+    class ParallelExecutor(object):
         """
         ParallelExecutor is designed for data parallelism, which focuses on distributing
         the data across different nodes and every node operates on the data in parallel.