Important bug fix related to MPI data type.

Sending size_t with MPI is dangerous as it can represent many types:
http:https://stackoverflow.com/questions/40807833/sending-size-t-type-data-with-mpi

So we use int for all dimension variables for safety.
Although the array size is thus less flexible.

NetCDF nc_inq_dimlen returns size_t. Simply convert it back to int.

Parallel_Algorithm/MPI/Kmean_mpi.c

@@ -26,7 +26,7 @@ int main() {
  MPI_Comm_size(MPI_COMM_WORLD, &size);
  //printf("hello world from process %d of %d\n", rank, size);
 
- size_t N_samples_all,N_samples,N_features,N_clusters,N_repeat;
+ int N_samples_all,N_samples,N_features,N_clusters,N_repeat;
  //i for samples; j for features; k for clusters (typically)
  int i,j,k;
  int k_best,initial_idx;
@@ -68,8 +68,7 @@ int main() {
  TBD: use MPI_Scatterv to handle arbitrary size
  */
 
- //convert to int to prevent error from unsigned/signed
- N_samples = (int)N_samples_all / size; 
+ N_samples = N_samples_all / size; 
  // printf("%d, Local samples: %d \n",rank,N_samples);
 
  if (rank==0){
@@ -82,8 +81,7 @@ int main() {
 
  // check scattered results
  if (rank==size-1){
- //convert to int to prevent error from unsigned - signed
- printf("Last element after sacattering %d: %f \n",rank,X[(int)N_samples-1][(int)N_features-1]);
+ printf("Last element after sacattering %d: %f \n",rank,X[N_samples-1][N_features-1]);
  }
 
  double iElaps1 = MPI_Wtime() - iStart1;
@@ -108,7 +106,6 @@ int main() {
  // how many data points in the cluster
  // needed by calculating the average position of data points in each cluster
  int* cluster_sizes = (int *)malloc(N_clusters*sizeof(int));
- // cluster_sizes[0] = 0; //why this will fail for rank1?
 
  /*
  ======================================================
@@ -138,9 +135,9 @@ int main() {
  if (rank==1){
  // initialize other nodes
  for (k=0; k<N_clusters; k++){
- //cluster_sizes[k] = 0; 
+ cluster_sizes[k] = 0; 
  for (j=0; j<N_features; j++){
- //new_cluster_centers[k][j] = 0.0;
+ new_cluster_centers[k][j] = 0.0;
  }
  }
  }

Parallel_Algorithm/OpenMP/Kmean_omp.c

@@ -18,7 +18,7 @@ int main() {
  ---------------- Initialization ---------------------
  ======================================================
  */
- size_t N_samples,N_features,N_clusters,N_repeat;
+ int N_samples,N_features,N_clusters,N_repeat;
  //i for samples; j for features; k for clusters (typically)
  int i,j,k;
  int k_best,initial_idx;

Parallel_Algorithm/shared/ncdf_util.c

@@ -20,10 +20,13 @@
  * (need to pass the address)
  */
 int readX(char* FILE_NAME, float*** p_X,int*** p_GUESS,
- size_t* p_N_samples,size_t* p_N_features,
- size_t* p_N_clusters,size_t* p_N_repeat ) {
+ int* p_N_samples,int* p_N_features,
+ int* p_N_clusters,int* p_N_repeat ) {
  int ncid, varid,dimid;
  int retval;
+ size_t N_temp;
+
+ printf("reading data \n");
 
  /* Open the file. NC_NOWRITE tells netCDF we want read-only access
  * to the file.*/
@@ -32,19 +35,23 @@ int readX(char* FILE_NAME, float*** p_X,int*** p_GUESS,
 
  /* Get the size of the data for dynamical allocation*/
  nc_inq_dimid(ncid,"N_samples",&dimid);
- nc_inq_dimlen(ncid,dimid,p_N_samples);
+ nc_inq_dimlen(ncid,dimid,&N_temp);
+ *p_N_samples = (int)N_temp;
  printf("Number of samples: %d \n",*p_N_samples);
 
  nc_inq_dimid(ncid,"N_features",&dimid);
- nc_inq_dimlen(ncid,dimid,p_N_features);
+ nc_inq_dimlen(ncid,dimid,&N_temp);
+ *p_N_features = (int)N_temp;
  printf("Number of features: %d \n",*p_N_features);
 
  nc_inq_dimid(ncid,"N_clusters",&dimid);
- nc_inq_dimlen(ncid,dimid,p_N_clusters);
+ nc_inq_dimlen(ncid,dimid,&N_temp);
+ *p_N_clusters = (int)N_temp;
  printf("Number of clusters: %d \n",*p_N_clusters);
 
  nc_inq_dimid(ncid,"N_repeat",&dimid);
- nc_inq_dimlen(ncid,dimid,p_N_repeat);
+ nc_inq_dimlen(ncid,dimid,&N_temp);
+ *p_N_repeat = (int)N_temp;
  printf("Number of repeated runs: %d \n",*p_N_repeat);
 
  /* Get the varid of the data variable, based on its name. */

Parallel_Algorithm/shared/ncdf_util.h

@@ -2,8 +2,8 @@
 #define NCDF_UTIL_H
 
 int readX(char* FILE_NAME, float*** p_X,int*** p_GUESS,
- size_t* p_N_samples,size_t* p_N_features,
- size_t* p_N_clusters,size_t* p_N_repeat );
+ int* p_N_samples,int* p_N_features,
+ int* p_N_clusters,int* p_N_repeat );
 
 int writeY(char* FILE_NAME, int* labels, float inert);