diff --git a/paddle/gserver/layers/MKLDNNConvLayer.cpp b/paddle/gserver/layers/MKLDNNConvLayer.cpp
index 617874defe74bd85b07c33de1c6bafdcadc768c4..19891043a1fa11c947118cf952171fec18294191 100644
--- a/paddle/gserver/layers/MKLDNNConvLayer.cpp
+++ b/paddle/gserver/layers/MKLDNNConvLayer.cpp
@@ -47,9 +47,9 @@ bool MKLDNNConvLayer::init(const LayerMap& layerMap,
   sw_ = conf.stride();
   sh_ = conf.stride_y();
   gp_ = conf.groups();
-  oh_ = conf.has_output_y() ? conf.output_y() : conf.output_x();
+  oh_ = conf.output_y();
   ow_ = conf.output_x();
-  ih_ = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
+  ih_ = conf.img_size_y();
   iw_ = conf.img_size();
   caffeMode_ = conf.caffe_mode();
   CHECK(caffeMode_) << "Only support caffe mode yet";
diff --git a/paddle/gserver/layers/MKLDNNConvLayer.h b/paddle/gserver/layers/MKLDNNConvLayer.h
index 58891ff5e1f5162dd1dadec9e45d72a6d0e5e1b9..d1a78ac1c0fd29f5e88511055c2b4e0d7cc91e88 100644
--- a/paddle/gserver/layers/MKLDNNConvLayer.h
+++ b/paddle/gserver/layers/MKLDNNConvLayer.h
@@ -37,38 +37,30 @@ protected:
   // group number
   int gp_;
 
-  // in backward data the format is different with wgtVal_
+  // in resetBwdData, the format of wgtValBwdData_ is different with wgtVal_
   MKLDNNMatrixPtr wgtValBwdData_;
+  // convert handle from wgtVal_ to wgtValBwdData_
   std::shared_ptr<mkldnn::reorder> cvtWgtVal_;
 
-  // save forward primitive_desc use for backward
+  // save forward primitive_desc, which can be used backward
   std::shared_ptr<mkldnn::convolution_forward::primitive_desc> fwdPD_;
 
-  // MKLDNNMatrixPtr with cpu device for conversion between MKLDNN device
+  // MKLDNNMatrixPtr which should be created from CPU Device
   MKLDNNMatrixPtr cpuInVal_;
   MKLDNNMatrixPtr cpuInGrad_;
   MKLDNNMatrixPtr cpuOutVal_;
   MKLDNNMatrixPtr cpuOutGrad_;
+  // convert handle between CPU device and MKLDNN device
   std::shared_ptr<mkldnn::reorder> cvtInVal_;
   std::shared_ptr<mkldnn::reorder> cvtInGrad_;
   std::shared_ptr<mkldnn::reorder> cvtOutVal_;
   std::shared_ptr<mkldnn::reorder> cvtOutGrad_;
 
-  // if has already init the weight
+  // whether the weight has been init
   bool hasInitedWgt_;
 
-  // True by default. This impact the calculation of output size.
-  // For example:
-  // - input(+padding): 0123456789
-  // - imageSize(+padding) = 10;
-  // - filterSize = 3;
-  // - stride = 2;
-  // - caffeMode_ is true:
-  // - output: (012), (234), (456), (678)
-  // - outputSize = 4;
-  // - caffeMode_ is false:
-  // - output: (012), (234), (456), (678), (9)
-  // - outputSize = 5;
+  // true by default, which impact the calculation of output image size.
+  // details can refer to mathUtil.h
   bool caffeMode_;
 
   // weight and bias
diff --git a/paddle/math/MKLDNNMatrix.cpp b/paddle/math/MKLDNNMatrix.cpp
index a71ac12afcea63bf2e543ff8ec55be6d33be8571..0778bb63b7b3bca9b3d2647ca43dad72d783950a 100644
--- a/paddle/math/MKLDNNMatrix.cpp
+++ b/paddle/math/MKLDNNMatrix.cpp
@@ -52,11 +52,7 @@ MKLDNNMatrixPtr MKLDNNMatrix::create(MatrixPtr m,
 std::shared_ptr<reorder> MKLDNNMatrix::createReorder(const MKLDNNMatrixPtr& src,
                                                      const MKLDNNMatrixPtr& dst,
                                                      bool checkData) {
-  if (src == dst) {
-    return nullptr;
-  }
-
-  if (src->getPrimitiveDesc() == dst->getPrimitiveDesc()) {
+  if (src == dst || src->getPrimitiveDesc() == dst->getPrimitiveDesc()) {
     return nullptr;
   }
 
diff --git a/paddle/math/MKLDNNMatrix.h b/paddle/math/MKLDNNMatrix.h
index c7765369c8571567a377e8603ab41850d80bba26..0aa130b4a0d458ad78d5d1330164af9e73b22a44 100644
--- a/paddle/math/MKLDNNMatrix.h
+++ b/paddle/math/MKLDNNMatrix.h
@@ -65,6 +65,12 @@ public:
 
   /**
    * Create reorder primitive.
+   * Create a mkldnn::reorder handle for converting src MKLDNNMatrix to dst.
+   * checkData: for whether to check the data handle of src and dst is the same.
+   *            if true, means check it and do not want support inplace reorder;
+   *            otherwise do not check data which means the created reorder
+   *            maybe inplace buffer and do not guarantee the logical is correct
+   *            since not all format or conversion support inplace.
    */
   static std::shared_ptr<mkldnn::reorder> createReorder(
       const MKLDNNMatrixPtr& src,
diff --git a/python/paddle/trainer/config_parser.py b/python/paddle/trainer/config_parser.py
index d633cae4aa178361adc376e31bcd4365599a702a..58ebcd1d8e01c6bcad5ef364709135a632e14c2d 100644
--- a/python/paddle/trainer/config_parser.py
+++ b/python/paddle/trainer/config_parser.py
@@ -2073,10 +2073,7 @@ class ConvLayerBase(LayerBase):
             (parallel_nn == 0 or self.config.device > -1)):
             self.layer_type = "cudnn_conv"
         else:
-            if (use_mkldnn == 1):
-                self.layer_type = "mkldnn_conv"
-            else:
-                self.layer_type = "exconv"
+            self.layer_type = "mkldnn_conv" if use_mkldnn else "exconv"
         # need to specify layer in config
         self.config.type = self.layer_type