feat: finish parallel merge sort for arbitrary process size

src/merge-parallel.c

@@ -2,6 +2,7 @@
 #include <omp.h>
 #include <stdio.h>
 #include <stdlib.h>
+#include <string.h>
 
 /*
  * Procedure merge_array
@@ -107,9 +108,9 @@ int main() {
 
 	MPI_Bcast(&arr_size, 1, MPI_INT, 0, MPI_COMM_WORLD);
 
-	// split array to processes
+	// -*- split array to processes -*-
 	int local_arr_size = arr_size / world_size + arr_size % world_size;
-	int local_arr[local_arr_size];
+	int local_arr[arr_size];
 	int *displs = NULL;
 	int *counts = NULL;
 
@@ -126,14 +127,12 @@ int main() {
 			counts[i] = local_arr_size - (i != 0 ? arr_size % world_size : 0);
 			displs[i] = next_displ;
 			next_displ = counts[i] + displs[i];
-
-			// printf("%d: counts %d\tdispls %d\n", i, counts[i], displs[i]);
 		}
-		// printf("\n");
 	} else {
 		local_arr_size -= arr_size % world_size;
 	}
 
+	// send the array to all nodes
 	MPI_Scatterv(
 		arr,
 		counts,
@@ -145,82 +144,97 @@ int main() {
 		0,
 		MPI_COMM_WORLD
 	);
+	// -*- finish splitting arrays -*-
 
-	// finish splitting arrays
-
-	// sort arrays
+	// sort local array
 	merge_sort(local_arr, 0, local_arr_size - 1);
 
-	// printf("before rank %d: ", my_rank);
-	// for (int i = 0; i < local_arr_size; ++i) {
-	// 	printf("%d, ", local_arr[i]);
-	// }
-	// printf("\n");
-
 	// send sort result to be merged
-	int recv_arr_sz = 0;
-	if (my_rank % 2 == 0) {
-		MPI_Recv(
-			&recv_arr_sz,
-			1,
-			MPI_INT,
-			my_rank + 1,
-			my_rank,
-			MPI_COMM_WORLD,
-			MPI_STATUS_IGNORE
-		);
-	} else {
-		MPI_Send(
-			&local_arr_size,
-			1,
-			MPI_INT,
-			my_rank - 1,
-			my_rank - 1,
-			MPI_COMM_WORLD
-		);
-	}
-
-	int recv_arr[recv_arr_sz];
-
-	if (my_rank % 2 == 0) {
-		MPI_Recv(
-			recv_arr,
-			recv_arr_sz,
-			MPI_INT,
-			my_rank + 1,
-			my_rank,
-			MPI_COMM_WORLD,
-			MPI_STATUS_IGNORE
-		);
-	} else {
-		MPI_Send(
-			local_arr,
-			local_arr_size,
-			MPI_INT,
-			my_rank - 1,
-			my_rank - 1,
-			MPI_COMM_WORLD
-		);
-	}
-
-	if (my_rank % 2 == 0) {
-		int total_sz = local_arr_size + recv_arr_sz;
-		int total_arr[local_arr_size + recv_arr_sz];
-
-		merge_two_arrays(local_arr, local_arr_size, recv_arr, recv_arr_sz, total_arr);
-
-		printf("after rank %d: ", my_rank);
-		for (int i = 0; i < total_sz; ++i) {
-			printf("%d, ", total_arr[i]);
+	for (int i = 2; i < world_size * 2 && my_rank % (i/2) == 0; i *= 2) {
+		int recv_arr_sz = 0;
+		int target_rank = my_rank + (my_rank % i == 0 ? 1 : -1) * i/2;
+
+		if (target_rank < world_size) {
+			// receive array
+			// start by sending or receiving the size of the array
+			if (my_rank % i == 0) {
+				MPI_Recv(
+					&recv_arr_sz,
+					1,
+					MPI_INT,
+					target_rank,
+					my_rank,
+					MPI_COMM_WORLD,
+					MPI_STATUS_IGNORE
+				);
+			} else {
+				MPI_Send(
+					&local_arr_size,
+					1,
+					MPI_INT,
+					target_rank,
+					target_rank,
+					MPI_COMM_WORLD
+				);
+			}
+
+			// create receive array buffer
+			int recv_arr[recv_arr_sz];
+			// now send or receive the array
+			if (my_rank % i == 0) {
+				MPI_Recv(
+					recv_arr,
+					recv_arr_sz,
+					MPI_INT,
+					target_rank,
+					my_rank,
+					MPI_COMM_WORLD,
+					MPI_STATUS_IGNORE
+				);
+			} else {
+				MPI_Send(
+					local_arr,
+					local_arr_size,
+					MPI_INT,
+					target_rank,
+					target_rank,
+					MPI_COMM_WORLD
+				);
+			}
+
+			// merge the local array and recieved array
+			if (my_rank % i == 0) {
+				// create output array
+				int total_sz = local_arr_size + recv_arr_sz;
+				int total_arr[local_arr_size + recv_arr_sz];
+
+				merge_two_arrays(
+					local_arr,
+					local_arr_size,
+
+					recv_arr,
+					recv_arr_sz,
+
+					total_arr
+				);
+
+				// copy output array to the local array
+				local_arr_size = total_sz;
+				memcpy(local_arr, total_arr, local_arr_size * sizeof(int));
+			}
 		}
-		printf("\n");
-
 	}
 
 	if (my_rank == 0) {
 		free(arr);
 		free(displs);
 		free(counts);
+
+		// output
+		for (int i = 0; i < local_arr_size; ++i) {
+			printf("%d, ", local_arr[i]);
+		}
+		printf("\n");
 	}
 
 	MPI_Finalize();