diff --git a/src/layer/arm/lrn_arm.cpp b/src/layer/arm/lrn_arm.cpp
index a5b5fb483..4f5b471b3 100644
--- a/src/layer/arm/lrn_arm.cpp
+++ b/src/layer/arm/lrn_arm.cpp
@@ -200,12 +200,14 @@ int LRN_arm::forward_inplace(Mat& bottom_top_blob) const
         for (int q=0; q<channels; q++)
         {
             float* ptr = bottom_top_blob.channel(q);
-            const float* sptr = square_blob_bordered.channel(q);
+            const Mat m = square_blob_bordered.channel(q);
 
             for (int i = 0; i < outh; i++)
             {
                 for (int j = 0; j < outw; j++)
                 {
+                    const float* sptr = m.row(i) + j;
+
                     float ss = 0.f;
 
                     for (int k = 0; k < maxk; k++)
@@ -218,7 +220,6 @@ int LRN_arm::forward_inplace(Mat& bottom_top_blob) const
                 }
 
                 ptr += outw;
-                sptr += w;
             }
         }
     }
diff --git a/src/layer/lrn.cpp b/src/layer/lrn.cpp
index f3c840695..4eea25a38 100644
--- a/src/layer/lrn.cpp
+++ b/src/layer/lrn.cpp
@@ -174,13 +174,15 @@ int LRN::forward(const Mat& bottom_blob, Mat& top_blob) const
         for (int q=0; q<channels; q++)
         {
             const float* ptr = bottom_blob.channel(q);
-            const float* sptr = square_blob_bordered.channel(q);
+            const Mat m = square_blob_bordered.channel(q);
             float* outptr = top_blob.channel(q);
 
             for (int i = 0; i < outh; i++)
             {
                 for (int j = 0; j < outw; j++)
                 {
+                    const float* sptr = m.row(i) + j;
+
                     float ss = 0.f;
 
                     for (int k = 0; k < maxk; k++)
@@ -193,7 +195,6 @@ int LRN::forward(const Mat& bottom_blob, Mat& top_blob) const
                 }
 
                 ptr += outw;
-                sptr += w;
                 outptr += outw;
             }
         }
@@ -305,12 +306,14 @@ int LRN::forward_inplace(Mat& bottom_top_blob) const
         for (int q=0; q<channels; q++)
         {
             float* ptr = bottom_top_blob.channel(q);
-            const float* sptr = square_blob_bordered.channel(q);
+            const Mat m = square_blob_bordered.channel(q);
 
             for (int i = 0; i < outh; i++)
             {
                 for (int j = 0; j < outw; j++)
                 {
+                    const float* sptr = m.row(i) + j;
+
                     float ss = 0.f;
 
                     for (int k = 0; k < maxk; k++)
@@ -323,7 +326,6 @@ int LRN::forward_inplace(Mat& bottom_top_blob) const
                 }
 
                 ptr += outw;
-                sptr += w;
             }
         }
     }
diff --git a/src/layer/pooling.cpp b/src/layer/pooling.cpp
index f4bd584d5..f79dbc51d 100644
--- a/src/layer/pooling.cpp
+++ b/src/layer/pooling.cpp
@@ -229,7 +229,7 @@ int Pooling::forward(const Mat& bottom_blob, Mat& top_blob) const
         #pragma omp parallel for
         for (int q=0; q<channels; q++)
         {
-            const Mat m(w, h, bottom_blob_bordered.channel(q));
+            const Mat m = bottom_blob_bordered.channel(q);
             float* outptr = top_blob.channel(q);
 
             for (int i = 0; i < outh; i++)
@@ -258,7 +258,7 @@ int Pooling::forward(const Mat& bottom_blob, Mat& top_blob) const
         #pragma omp parallel for
         for (int q=0; q<channels; q++)
         {
-            const Mat m(w, h, bottom_blob_bordered.channel(q));
+            const Mat m = bottom_blob_bordered.channel(q);
             float* outptr = top_blob.channel(q);
 
             for (int i = 0; i < outh; i++)