implemented point exchange still preliminarly

src/tree/tree.c

@@ -958,7 +958,7 @@ void build_top_kdtree(global_context_t *ctx, pointset_t *og_pointset, top_kdtree
 	MPI_DB_PRINT("Root is %p\n", tree -> root);
 	if(I_AM_MASTER)
 	{
-		tree_print(ctx, tree -> root);
+		//tree_print(ctx, tree -> root);
 		write_nodes_to_file(ctx, tree, "bb/nodes_50_blobs_more_var.csv");
 	}
 
@@ -1031,8 +1031,9 @@ int partition_data_around_key(int* key, float_t *val, int vec_len, int ref_key ,
 
 void exchange_points(global_context_t* ctx, top_kdtree_t* tree)
 {
-	size_t* points_per_proc = (size_t*)malloc(ctx -> world_size * sizeof(size_t));	
+	int* points_per_proc = (int*)malloc(ctx -> world_size * sizeof(int));	
 	int* points_owners 	 = (int*)malloc(ctx -> dims * ctx -> local_n_points * sizeof(float_t));
+	int* partition_offset = (int*)malloc(ctx -> world_size * sizeof(int));	
 	/* compute owner */
 	for(size_t i = 0; i < ctx -> local_n_points; ++i)
 	{
@@ -1044,9 +1045,12 @@ void exchange_points(global_context_t* ctx, top_kdtree_t* tree)
 	int last_idx = 0;
 	int len 	 = ctx -> local_n_points;
 	float_t* curr_data = ctx -> local_data;
+
+	partition_offset[0] = 0;
 	for(int owner = 1; owner < ctx -> world_size; ++owner)
 	{
 		last_idx = partition_data_around_key(points_owners, ctx -> local_data, ctx -> dims, owner, last_idx, ctx -> local_n_points);	
+		partition_offset[owner] = last_idx;
 		points_per_proc[owner - 1] = last_idx;
 	}
 
@@ -1060,39 +1064,68 @@ void exchange_points(global_context_t* ctx, top_kdtree_t* tree)
 
 
 	/*
-	MPI_DB_PRINT("Points per proc begin: ");
-	for(int i = 0; i < ctx -> world_size; ++i)
-	{
-		MPI_DB_PRINT("%lu ", points_per_proc[i]);
-	}
-	MPI_DB_PRINT("\n");
-	*/
-
-	MPI_Allreduce(MPI_IN_PLACE, points_per_proc, ctx -> world_size, MPI_UINT64_T,MPI_SUM, ctx -> mpi_communicator);
+	int* points_per_proc_all = (int*)malloc(ctx -> world_size * sizeof(int));	
+	MPI_Allreduce(MPI_IN_PLACE, points_per_proc_all, ctx -> world_size, MPI_INT,MPI_SUM, ctx -> mpi_communicator);
 
 	size_t test_num = 0;
-	for(int i = 0; i < ctx -> world_size; ++i) test_num += points_per_proc[i];
+	for(int i = 0; i < ctx -> world_size; ++i) test_num += points_per_proc_all[i];
 	MPI_DB_PRINT("Master has n_points %lu and in node population has %lu points\n", ctx -> n_points, test_num);
-	/*
+
+
+	
 	MPI_DB_PRINT("Points per proc after: ");
 	for(int i = 0; i < ctx -> world_size; ++i)
 	{
-		MPI_DB_PRINT("%lu ", points_per_proc[i]);
+		MPI_DB_PRINT("%lu ", points_per_proc_all[i]);
 	}
 	MPI_DB_PRINT("\n");
+
+	free(points_per_proc_all);
 	*/
 
+	int* 	 rcvcount  = (int*)malloc(ctx -> world_size * sizeof(int));
+	int* 	 displs    = (int*)malloc(ctx -> world_size * sizeof(int));
+	float_t* rcvbuffer = NULL;
+	int tot_count = 0;
+	for(int rcv = 0; rcv < ctx -> world_size; ++rcv)
+	{
+		/* recieve the number of points to recieve from each proc */
+		MPI_Gather(&(points_per_proc[rcv]), 1, MPI_INT, rcvcount, 1, MPI_INT, rcv, ctx -> mpi_communicator);
+		float_t* send_buffer = ctx -> local_data + (ctx -> dims * partition_offset[rcv]);
+
 
-	/*
-	for(int i = 0; i < ctx -> local_n_points; ++i)
+		/* if I am the reciever recieve */
+		if(rcv == ctx -> mpi_rank)
 		{
-		MPI_DB_PRINT("%d ", points_owners[i]); 
-		if(i % 10 == 0) MPI_DB_PRINT("\n");
+			displs[0] = 0;
+			for(int i = 1; i < ctx -> world_size; ++i) displs[i] = displs[i - 1] + rcvcount[i - 1];
+
+			/*multiply for number of elements */
+			for(int i = 0; i < ctx -> world_size; ++i) 
+			{
+				displs[i]   = displs[i] * ctx -> dims;
+				rcvcount[i] = rcvcount[i] * ctx -> dims;
+				tot_count += rcvcount[i];
 			}
-	*/
+			//DB_PRINT("[RANK %d] is recieving %d elements %d points\n", rcv, tot_count, tot_count / ctx -> dims);
+			rcvbuffer = (float_t*)malloc(tot_count * sizeof(float_t));
+		}
+		MPI_Gatherv(send_buffer, points_per_proc[rcv], MPI_MY_FLOAT, rcvbuffer, rcvcount, displs, MPI_MY_FLOAT, rcv, ctx -> mpi_communicator);			
+	}
+
+	ctx -> local_n_points = tot_count; 
+
+	/* free prv pointer */
+	free(ctx -> local_data);
+	ctx -> local_data = rcvbuffer;
+	
+
 
 	free(points_owners);
 	free(points_per_proc);
+	free(partition_offset);
+	free(rcvcount);
+	free(displs);
 }
 
 void simulate_master_read_and_scatter(int dims, size_t n, global_context_t *ctx) 
@@ -1184,7 +1217,6 @@ void simulate_master_read_and_scatter(int dims, size_t n, global_context_t *ctx)
 	original_ps.lb_box = (float_t*)malloc(ctx -> dims * sizeof(float_t));
 	original_ps.ub_box = (float_t*)malloc(ctx -> dims * sizeof(float_t));
 
-	float_t incr = 0.05;
 	float_t tol = 0.002;
 
 	top_kdtree_t tree;
@@ -1196,6 +1228,7 @@ void simulate_master_read_and_scatter(int dims, size_t n, global_context_t *ctx)
 
 
 
+
 	free(send_counts);
 	free(displacements);