Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into ini

Dockerfile

 # A image for building paddle binaries
 # Use cuda devel base image for both cpu and gpu environment
-FROM nvidia/cuda:8.0-cudnn5-devel-ubuntu16.04
+FROM nvidia/cuda:8.0-cudnn7-devel-ubuntu16.04
 MAINTAINER PaddlePaddle Authors <paddle-dev@baidu.com>
 
 ARG UBUNTU_MIRROR

paddle/fluid/operators/CMakeLists.txt

@@ -245,9 +245,17 @@ op_library(channel_send_op DEPS concurrency)
 op_library(channel_recv_op DEPS concurrency)
 
 list(REMOVE_ITEM GENERAL_OPS ${DEPS_OPS})
+
+# The fully connected layer is deleted when the WITH_MKLDNN flag is OFF
+# Because the fully connected layer has only one MKLDNN's operator
+if(NOT WITH_MKLDNN)
+    list(REMOVE_ITEM GENERAL_OPS fc_op)
+endif(NOT WITH_MKLDNN)
+
 foreach(src ${GENERAL_OPS})
     op_library(${src})
 endforeach()
+
 file(APPEND ${pybind_file} "USE_OP(less_than);\nUSE_OP(logical_and);\nUSE_NO_KERNEL_OP(read_from_array);\n")
 
 add_subdirectory(reader)

paddle/fluid/operators/reshape_op.h

@@ -60,7 +60,7 @@ class ReshapeOp : public framework::OperatorWithKernel {
   static framework::DDim ValidateShape(const std::vector<int> shape,
                                        const framework::DDim &in_dims) {
     const int64_t in_size = framework::product(in_dims);
-    // only one dimension canbe set to -1, whose size will be automatically
+    // only one dimension can be set to -1, whose size will be automatically
     // infered.
     const int64_t unk_dim_val = -1;
     const int64_t copy_dim_val = 0;
@@ -119,13 +119,15 @@ class ReshapeKernel : public framework::OpKernel<T> {
     auto *shape_tensor = ctx.Input<framework::LoDTensor>("Shape");
 
     framework::DDim out_dims = out->dims();
+
     if (shape_tensor) {
       auto *shape_data = shape_tensor->data<int>();
-      if (platform::is_gpu_place(ctx.GetPlace())) {
       framework::Tensor cpu_shape_tensor;
+      if (platform::is_gpu_place(ctx.GetPlace())) {
         TensorCopy(*shape_tensor, platform::CPUPlace(), ctx.device_context(),
                    &cpu_shape_tensor);
         shape_data = cpu_shape_tensor.data<int>();
+        ctx.device_context().Wait();
       }
       auto shape =
           std::vector<int>(shape_data, shape_data + shape_tensor->numel());

python/paddle/fluid/tests/unittests/CMakeLists.txt

 file(GLOB TEST_OPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "test_*.py")
 string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}")
 
+# The fully connected test is removed whe the WITH_MKLDNN flag is OFF
+# Because the fully connected layer has only one kernel (MKLDNN)
+if(NOT WITH_MKLDNN)
+    list(REMOVE_ITEM TEST_OPS test_fc_op)
+endif(NOT WITH_MKLDNN)
+
 if(NOT WITH_DISTRIBUTE)
     list(REMOVE_ITEM TEST_OPS test_recv_op)
 endif(NOT WITH_DISTRIBUTE)