diff --git a/allvars.h b/allvars.h
index 7610de5555887b11c955280e10ca783e073ae6ef..2f66fc12b14ddf063a573cbf284f6aab6f4bc02b 100644
--- a/allvars.h
+++ b/allvars.h
@@ -30,8 +30,6 @@
 #include "numa.h"
 
 
-#include "numa.h"
-
 extern struct io
 {
 	FILE * pFile;
diff --git a/gridding.c b/gridding.c
index 1e34d604495b9398fe3e00d1af214e39353128b8..76d1e88bd80255e5fc16b495458c51d0512dc14c 100644
--- a/gridding.c
+++ b/gridding.c
@@ -1,4 +1,5 @@
 #include <stdio.h>
+#include <math.h>
 #include "allvars.h"
 #include "proto.h"
 
@@ -253,10 +254,12 @@ void gridding_data(){
       // MPI_Win_unlock(target_rank,slabwin);
        memcpy(Me.win.ptr+isector*sizeof(gridss), gridss, sizeof(gridss));
        MPI_Barrier(MPI_COMM_WORLD);
-       printf("rank %d\n", rank);
-       reduce(target_rank, isector, 0, size-1);
+       if(Me.Rank[myHOST] == 0)
+       	  reduce(isector, 1<<(++Me.Ntasks[myHOST]));
        if((Me.MAXl > myHOST) && (Me.Rank[HOSTS] != -1))
-         MPI_Reduce(grid,grid,size_of_grid,MPI_DOUBLE,MPI_SUM,target_rank,*Me.COMM[HOSTS]);  
+          MPI_Reduce(grid,grid,size_of_grid,MPI_DOUBLE,MPI_SUM,target_rank,*Me.COMM[HOSTS]);
+       else
+          memcpy(grid, Me.swins[Me.Rank[myHOST]].ptr+isector*sizeof(gridss), sizeof(grid));   
        //MPI_Put(gridss,size_of_grid,MPI_DOUBLE,target_rank,0,size_of_grid,MPI_DOUBLE,slabwin);
       #else
        MPI_Reduce(gridss,grid,size_of_grid,MPI_DOUBLE,MPI_SUM,target_rank,MPI_COMM_WORLD);
@@ -384,32 +387,21 @@ void write_grided_data()
 
 }
 
-void reduce(int target, int sector, int begin, int end)
+void reduce(int sector, int max_level)
 {
-   if(begin >= end)
-       return;
-   int mid = (begin+end)/2;
-   reduce(target, sector, begin, mid);
-   reduce(target, sector, mid+1, end);
-   accumulate(target, sector, begin, mid, end);
+   for(int i = 0; i < max_level; i++)
+   {
+         int threshold = 1 << (i+1);
+         if( Me.Rank[myHOST] % threshold == 0)
+         {
+         	int local = Me.Rank[myHOST]+ (1<<i);
+         	for(int j=0; j<size_of_grid; j++)
+         	{
+	 		*((double*)Me.swins[Me.Rank[myHOST]].ptr+sector*sizeof(gridss)+j) += *((double*)Me.swins[local].ptr+sector*sizeof(gridss)+j);
+         	}  
+         }
+   }
 }
 
 
-void accumulate(int target, int sector, int begin, int mid, int end)
-{
-
-    for(int i =mid; i<end; i++)
-    {
-        if(i != target)
-        {
-             for(int j=0; j<size_of_grid; j++)
-             {
-                    *((double*)Me.swins[target].ptr+sector*sizeof(gridss)+j) += *((double*)Me.swins[i].ptr+sector*sizeof(gridss)+j);
-                    //printf("value = %lf, rank = %d\n", *((double*)Me.swins[target].ptr+isector*sizeof(gridss)+j), rank);
-             }
-        }
-     }
-    //if(rank == target_rank)
-    //   memcpy(grid, Me.swins[target_rank].ptr+isector*sizeof(gridss), sizeof(grid));    
 
-}
diff --git a/proto.h b/proto.h
index 353b6a79ed1e3e56d2ef4666774f6109a2b66389..d4618d533570d7281ab4a52aa02787da939c9dbd 100644
--- a/proto.h
+++ b/proto.h
@@ -23,8 +23,7 @@ void gridding();
 void initialize_array();
 void gridding_data();
 void write_grided_data();
-void reduce(int target, int sector, int begin, int end);
-void accumulate(int target, int sector, int begin, int mid, int end);
+void reduce(int sector, int max_level);
 
 
 /* fourier_transform.c */