diff --git a/paddle/function/GemmConvOp.cpp b/paddle/function/GemmConvOp.cpp
index 25cc3df6672d4f08383266e602bebd009e422bac..cbdbf5335d32d55a0221728758025c9d2cb3e7d1 100644
--- a/paddle/function/GemmConvOp.cpp
+++ b/paddle/function/GemmConvOp.cpp
@@ -189,8 +189,8 @@ public:
     size_t colHeight = inputChannels / groups_ * filterHeight * filterWidth;
     size_t colWidth = outputHeight * outputWidth;
     // Max col matrix height 256, Max col matrix width 1024
-    size_t stepColHeight = std::min(colHeight, (size_t)256);
-    size_t stepColWidth = std::min(colWidth, (size_t)2048);
+    size_t stepColHeight = std::min(colHeight, static_cast<size_t>(256));
+    size_t stepColWidth = std::min(colWidth, static_cast<size_t>(2048));
 
     if (needIm2col) {
       colShape = TensorShape({inputChannels / groups_,
@@ -278,6 +278,8 @@ public:
       inputData += inputChannels * inputHeight * inputWidth;
       outputData += outputChannels * outputHeight * outputWidth;
     }
+
+    memory_.reset();
   }
 };
 
diff --git a/paddle/function/Im2Col.h b/paddle/function/Im2Col.h
index 1053e4fd232b8803c9661ece823da88715ffc220..36a9bcf84e4b14965c83627821b71d1c7c0da1b2 100644
--- a/paddle/function/Im2Col.h
+++ b/paddle/function/Im2Col.h
@@ -136,7 +136,7 @@ public:
             (imRowIdx - paddingHeight) >= inputHeight ||
             (imColIdx - paddingWidth) < 0 ||
             (imColIdx - paddingWidth) >= inputWidth) {
-          colData[colh * colWidthSize + colw] = T(0);
+          colData[colh * colWidthSize + colw] = static_cast<T>(0);
         } else {
           imRowIdx += c_im * inputHeight - paddingHeight;
           imColIdx -= paddingWidth;
diff --git a/paddle/function/Im2ColTest.cpp b/paddle/function/Im2ColTest.cpp
index c573469168d3c18cfc574570c0681beb60a52693..3ba866dcdd845403d52f7a85adfef08cbb11c305 100644
--- a/paddle/function/Im2ColTest.cpp
+++ b/paddle/function/Im2ColTest.cpp
@@ -140,13 +140,13 @@ TEST(Im2ColFunctor, GPU) { TestIm2ColFunctor<DEVICE_TYPE_GPU, float>(); }
 
 template <class T>
 void TestIm2ColMobileFunctor() {
-  for (size_t channels : {1, 5, 32}) {
-    for (size_t inputHeight : {5, 33, 100}) {
-      for (size_t inputWidth : {5, 32, 96}) {
-        for (size_t filterHeight : {1, 5}) {
-          for (size_t filterWidth : {3, 7}) {
-            for (size_t stride : {1, 2}) {
-              for (size_t padding : {0, 1}) {
+  for (size_t channels : {32}) {
+    for (size_t inputHeight : {33, 100}) {
+      for (size_t inputWidth : {32, 96}) {
+        for (size_t filterHeight : {5}) {
+          for (size_t filterWidth : {7}) {
+            for (size_t stride : {2}) {
+              for (size_t padding : {1}) {
                 for (size_t dilation : {1, 3}) {
                   size_t filterSizeH = (filterHeight - 1) * dilation + 1;
                   size_t filterSizeW = (filterWidth - 1) * dilation + 1;