diff --git a/include/gdf/cffi/functions.h b/include/gdf/cffi/functions.h
index 77b671f1..590def44 100644
--- a/include/gdf/cffi/functions.h
+++ b/include/gdf/cffi/functions.h
@@ -884,3 +884,11 @@ gdf_error gdf_quantile_aprrox(	gdf_column*  col_in,       //input column;
                                 double       q,            //requested quantile in [0,1]
                                 void*        t_erased_res, //type-erased result of same type as column;
                                 gdf_context* ctxt);        //context info
+
+
+
+gdf_error gdf_order_by_asc_desc(
+		gdf_column * input_columns, //pointers to pointers of input columns
+		size_t num_inputs, //number of pointeres in the first parameter (e.g. number of columsn to sort by
+		gdf_column * output_indices, //a gdf_column that is pre allocated for storing sorted indices
+		gdf_valid_type * asc_desc_bitmask); //asc_desc bitmask e.g. 101 would mean sort the first and last oclumns ascending and the second one descending
diff --git a/include/sqls_rtti_comp.hpp b/include/sqls_rtti_comp.hpp
index 8b641276..c4f73afd 100644
--- a/include/sqls_rtti_comp.hpp
+++ b/include/sqls_rtti_comp.hpp
@@ -24,6 +24,8 @@
 #include <thrust/advance.h>
 #include <thrust/gather.h>
 
+#include "gdf/utils.h"
+
 //for int<n>_t:
 //
 #include <cstdint>
@@ -49,7 +51,8 @@ struct LesserRTTI
     columns_(cols),
     rtti_(types),
     sz_(sz),
-    vals_(nullptr)
+    vals_(nullptr),
+    asc_desc_bitmask_(nullptr)
   {
   }
 
@@ -61,11 +64,65 @@ struct LesserRTTI
     columns_(cols),
     rtti_(types),
     sz_(sz),
-    vals_(vals)
+    vals_(vals),
+    asc_desc_bitmask_(nullptr)
   {
   }
 
+  __host__ __device__
+  LesserRTTI(void* const* cols,
+	     int* const types,
+	     size_t sz,
+	     gdf_valid_type* const asc_desc_bitmask):
+    columns_(cols),
+    rtti_(types),
+    sz_(sz),
+    vals_(nullptr),
+    asc_desc_bitmask_(asc_desc_bitmask)
+  {
+  }
   
+/**
+ * Should be used when you want to sort multiple columns using asc / desc flags for each column
+ *
+ *
+ */
+  __host__ __device__
+   bool asc_desc_comparison(IndexT row1, IndexT row2) const
+   {
+     for(size_t col_index = 0; col_index < sz_; ++col_index)
+     {
+       gdf_dtype col_type = static_cast<gdf_dtype>(rtti_[col_index]);
+       bool asc;
+
+       if(asc_desc_bitmask_ == nullptr){
+    	   asc = true;
+       }else{
+    	   asc = gdf_is_valid(asc_desc_bitmask_, col_index);
+       }
+       //if flag == true
+
+       State state;
+       if(asc){
+           OpLess less(row1, row2);
+           state =type_dispatcher(less, col_type, col_index);
+       }else{
+           OpGreater greater(row1, row2);
+           state =type_dispatcher(greater, col_type, col_index);
+       }
+
+       switch( state )
+       {
+       case State::False:
+         return false;
+       case State::True:
+         return true;
+       case State::Undecided:
+         break;
+       }
+     }
+     return false;
+   }
 
   __host__ __device__
   bool equal(IndexT row1, IndexT row2) const
@@ -187,6 +244,36 @@ struct LesserRTTI
     IndexT row2_;
   };
 
+  struct OpGreater
+    {
+       __host__ __device__
+       OpGreater(IndexT row1, IndexT row2):
+         row1_(row1),
+         row2_(row2)
+      {
+      }
+
+       template<typename ColType>
+       __host__ __device__
+       State operator() (int col_index,
+  		       const void* const * columns,
+  		       ColType )
+      {
+        ColType res1 = LesserRTTI::at<ColType>(col_index, row1_, columns);
+        ColType res2 = LesserRTTI::at<ColType>(col_index, row2_, columns);
+
+        if( res1 > res2 )
+  	return State::True;
+        else if( res1 == res2 )
+  	return State::Undecided;
+        else
+  	return State::False;
+      }
+    private:
+      IndexT row1_;
+      IndexT row2_;
+    };
+
   struct OpEqual
   {
      __host__ __device__
@@ -336,6 +423,7 @@ struct LesserRTTI
   const int* const rtti_;
   size_t sz_;
   const void* const * vals_; //for filtering
+  const gdf_valid_type* asc_desc_bitmask_; //a bitmask that allows us to know whether or not a column should be sorted ascending or descending
 };
 
 //###########################################################################
@@ -379,6 +467,8 @@ void multi_col_order_by(size_t nrows,
 		 });
 }
 
+
+
 //###########################################################################
 //#                          Multi-column Filter:                           #
 //###########################################################################
@@ -709,4 +799,27 @@ size_t multi_col_group_by_avg_sort(size_t         nrows,
   return new_sz;
 }
 
+#include <iostream>
+template<typename IndexT>
+void multi_col_order_by_asc_desc(
+		void* const * d_col_data,
+		int* d_col_types,
+		size_t num_inputs,
+		gdf_valid_type * asc_desc_bitmask,
+		IndexT*      d_indx,
+		size_t nrows,
+		cudaStream_t   stream = NULL){
+
+	LesserRTTI<IndexT> f(d_col_data, d_col_types, num_inputs,asc_desc_bitmask);
+	std::cout<<"about to sequence"<<std::endl;
+	thrust::sequence(thrust::cuda::par.on(stream), d_indx, d_indx+nrows, 0);
+	std::cout<<"sequenced"<<std::endl;
+	thrust::sort(thrust::cuda::par.on(stream),
+			d_indx, d_indx+nrows,
+			[f] __host__ __device__ (IndexT i1, IndexT i2){
+		return f.asc_desc_comparison(i1, i2);
+	});
+
+	std::cout<<"sorted"<<std::endl;
+}
 
diff --git a/src/sqls_ops.cu b/src/sqls_ops.cu
index 2d5307fc..2199df5f 100644
--- a/src/sqls_ops.cu
+++ b/src/sqls_ops.cu
@@ -38,6 +38,23 @@ namespace{ //annonymus
     cudaMemcpy(d_types, h_types, ncols*sizeof(int), cudaMemcpyHostToDevice);//TODO: add streams
   }
 
+  //does not assume the columns were contiguous in memory but arer rather pointers to columns
+  void soa_col_info_arr(gdf_column** cols, size_t ncols, void** d_cols, int* d_types)
+  {
+    std::vector<void*> v_cols(ncols,nullptr);
+    std::vector<int>   v_types(ncols, 0);
+    for(size_t i=0;i<ncols;++i)
+      {
+        v_cols[i] = cols[i]->data;
+        v_types[i] = cols[i]->dtype;
+      }
+
+    void** h_cols = &v_cols[0];
+    int* h_types = &v_types[0];
+    cudaMemcpy(d_cols, h_cols, ncols*sizeof(void*), cudaMemcpyHostToDevice);//TODO: add streams
+    cudaMemcpy(d_types, h_types, ncols*sizeof(int), cudaMemcpyHostToDevice);//TODO: add streams
+  }
+
   template<typename T>
     using Vector = thrust::device_vector<T>;
 
@@ -1375,3 +1392,41 @@ gdf_error gdf_group_by_count(int ncols,                    // # columns
 }
 
 
+gdf_error gdf_order_by_asc_desc(
+		gdf_column * input_columns,
+		size_t num_inputs,
+		gdf_column * output_indices,
+		gdf_valid_type * asc_desc_bitmask){
+
+	std::cout<<"i mean this is the first fucking line...."<<std::endl;
+	//TODO: don't assume type of output is size_t
+	typedef size_t IndexT;
+	//TODO: make these allocations happen with the new memory manager when we can
+	//also we are kind of assuming they will just work, yeesh!
+	thrust::device_vector<size_t> test(2);
+
+	std::cout<<"can do basic shit...."<<std::endl;
+	thrust::device_vector<void*> d_cols(num_inputs);
+	thrust::device_vector<int>   d_types(num_inputs, 0);
+
+	void** d_col_data = d_cols.data().get();
+	int* d_col_types = d_types.data().get();
+
+	std::cout<<"about to run soa"<<std::endl;
+
+	soa_col_info(input_columns, num_inputs,  d_col_data, d_col_types);
+
+	std::cout<<"about to run mul col order by"<<std::endl;
+	multi_col_order_by_asc_desc(
+			d_col_data,
+			d_col_types,
+			num_inputs,
+			asc_desc_bitmask,
+			(size_t *) output_indices->data,
+			input_columns[0].size);
+
+	return GDF_SUCCESS;
+
+}
+
+