diff --git a/.gitmodules b/.gitmodules
index 836d824a6f5a..b0bf4a300582 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -1,6 +1,6 @@
 [submodule "3rdparty/mshadow"]
 	path = 3rdparty/mshadow
-	url = https://github.com/dmlc/mshadow.git
+	url = https://github.com/ZhennanQin/mshadow.git 
 [submodule "3rdparty/dmlc-core"]
 	path = 3rdparty/dmlc-core
 	url = https://github.com/dmlc/dmlc-core.git
diff --git a/3rdparty/mshadow b/3rdparty/mshadow
index 6dc04f7c729c..40d0a53b67e8 160000
--- a/3rdparty/mshadow
+++ b/3rdparty/mshadow
@@ -1 +1 @@
-Subproject commit 6dc04f7c729cd5c6c6210d5d4d2026a26ce0bfbf
+Subproject commit 40d0a53b67e88f346cefe29f7351bae451b3eae9
diff --git a/src/operator/convolution_v1-inl.h b/src/operator/convolution_v1-inl.h
index 758ce12d8006..e0baa4efb1d3 100644
--- a/src/operator/convolution_v1-inl.h
+++ b/src/operator/convolution_v1-inl.h
@@ -335,10 +335,11 @@ class ConvolutionV1Op : public Operator {
                                      oshape[2] * oshape[3]);
     // param_.workspace is in elements of sizeof(DType)
     // if param_.workspace is set to zero the nstep_ equals ishape[0] (batch)
-    nstep_ = std::max<index_t>(
-        std::min(static_cast<index_t>(param_.workspace) /
-          (shape_colunit_.Size() + shape_dstunit_.Size()), ishape[0]),
-      1);
+    nstep_ =
+        std::max<index_t>(std::min(static_cast<index_t>(param_.workspace / (shape_colunit_.Size() +
+                                                                            shape_dstunit_.Size())),
+                                   ishape[0]),
+                          1);
 
     mshadow::Shape<2> scol = mshadow::Shape2(shape_colunit_[0],
                                              shape_colunit_[1] * nstep_);
diff --git a/src/operator/nn/deconvolution-inl.h b/src/operator/nn/deconvolution-inl.h
index d89a489c0183..8f1470b9eb67 100644
--- a/src/operator/nn/deconvolution-inl.h
+++ b/src/operator/nn/deconvolution-inl.h
@@ -459,10 +459,11 @@ class DeconvolutionOp {
                                      oshape[1] / param_.num_group,
                                      oshape[2] * oshape[3]);
     // See convolution for workspace calculations. nstep_ will be the effective batch size
-    nstep_ = std::max<index_t>(
-        std::min(static_cast<index_t>(param_.workspace) /
-          (shape_colunit_.Size() + shape_dstunit_.Size()), ishape[0]),
-      1);
+    nstep_ =
+        std::max<index_t>(std::min(static_cast<index_t>(param_.workspace / (shape_colunit_.Size() +
+                                                                            shape_dstunit_.Size())),
+                                   ishape[0]),
+                          1);
 
     mshadow::Shape<2> scol = mshadow::Shape2(shape_colunit_[0],
                                              shape_colunit_[1] * nstep_);