files for lecture19

L19/extreme.c

+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <time.h>
+#include "vec.h"
+
+int main (int argc, char* argv[]) {
+
+    // read the filename from the command line
+    if (argc < 2) {
+	printf ("command usage: %s %s\n",argv[0],"filename");
+	return 1;
+    }
+    char* filename = argv[1];
+
+    // open the text file for reading
+    FILE* fptr;
+    fptr = fopen(filename,"r");
+
+    // need to check for null
+    if (fptr == 0) {
+	printf ("Error opening data file %s.\n",filename);
+	exit(1);
+    }
+
+    // read the number of points and the dimension of each point
+    int num_points, dim;
+    if (fscanf(fptr,"%*c %d %d",&num_points, &dim) != 2) {
+        printf ("error reading the number of points and the dimension\n");
+        return 1;
+    }
+
+    // Read vectors from stdin and store them in a 2d array
+    double* data = (double*)malloc(num_points*dim*sizeof(double));
+    if (data == NULL) {
+        printf ("malloc return NULL pointer!\n");
+        return 1;
+    }
+    for (int i=0;i<num_points;i++) {
+        if (vec_read_file(fptr,data+i*dim,dim) != dim) {
+            printf ("error reading the next point from the file %s\n",filename);
+            return 1;
+        }
+    }
+
+    // close the data file
+    fclose(fptr);
+
+    // start the timer 
+    clock_t start_time = clock();
+
+    // find the extreme pair
+    double max_dist_sq = 0;
+    int extreme[2];
+    int pairs_checked = 0;    
+    for (int i=0;i<num_points-1;i++) {
+        for (int j=i+1;j<num_points;j++) {
+            double dist_sq = vec_dist_sq(data+i*dim,data+j*dim,dim);
+	    pairs_checked += 1;	    
+            if (dist_sq > max_dist_sq) {
+                max_dist_sq = dist_sq;
+                extreme[0] = i;
+                extreme[1] = j;
+            }
+        }
+    }
+
+    // stop the timer and calculate elapsed time
+    clock_t end_time = clock();
+    double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
+
+    // output the results
+    printf ("elapsed time = %.4f seconds\n",elapsed_time);
+    printf ("pairs checked = %d\n",pairs_checked);    
+    printf ("Extreme Distance = %.2f\n",sqrt(max_dist_sq));
+    printf ("Extreme Pair = %d %d\n",extreme[0],extreme[1]);
+
+    // free memory allocated for dataset 
+    free(data);
+}

L19/instructions.txt

+interact -A cmda3634_rjh -p normal_q -t 5 --ntasks-per-node=4
+module load matplotlib
+mpicc -o mpi_sum mpi_sum.c
+mpiexec -n 4 ./mpi_sum 100000000

L19/mpi_extreme.c

+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <mpi.h>
+#include "vec.h"
+
+int main (int argc, char* argv[]) {
+
+    MPI_Init (&argc, &argv);
+
+    // MPI_COMM_WORLD is the default communicator that contains all ranks
+    int rank, size;
+    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+    MPI_Comm_size(MPI_COMM_WORLD, &size);
+    char node_name[MPI_MAX_PROCESSOR_NAME];
+    int node_name_len;
+    MPI_Get_processor_name(node_name,&node_name_len);
+
+    // make sure we are not running on a login node!
+    if ((strcmp(node_name,"tinkercliffs1") == 0) || 
+            (strcmp(node_name,"tinkercliffs2") == 0)) {
+        printf ("error : running on login node %s!\n",node_name);
+        return 1;
+    }
+
+    // read the filename from the command line
+    if (argc < 2) {
+	printf ("command usage: %s %s\n",argv[0],"filename");
+	return 1;
+    }
+    char* filename = argv[1];
+
+    // open the text file for reading
+    FILE* fptr;
+    fptr = fopen(filename,"r");
+
+    // need to check for null
+    if (fptr == 0) {
+	printf ("Error opening data file %s.\n",filename);
+	exit(1);
+    }
+
+    // read the number of points and the dimension of each point
+    int num_points, dim;
+    if (fscanf(fptr,"%*c %d %d",&num_points, &dim) != 2) {
+        printf ("error reading the number of points and the dimension\n");
+        return 1;
+    }
+
+    // Read vectors from stdin and store them in a 2d array
+    double* data = (double*)malloc(num_points*dim*sizeof(double));
+    if (data == NULL) {
+        printf ("malloc return NULL pointer!\n");
+        return 1;
+    }
+    for (int i=0;i<num_points;i++) {
+        if (vec_read_file(fptr,data+i*dim,dim) != dim) {
+            printf ("error reading the next point from the file %s\n",filename);
+            return 1;
+        }
+    }
+
+    // close the data file
+    fclose(fptr);
+
+    // start the timer
+    double start_time, end_time;
+    start_time = MPI_Wtime();
+
+    // find the extreme pair
+    double max_dist_sq = 0;
+    int extreme[2];
+    int pairs_checked = 0;    
+    for (int i=0;i<num_points-1;i++) {
+        for (int j=i+1;j<num_points;j++) {
+            double dist_sq = vec_dist_sq(data+i*dim,data+j*dim,dim);
+	    pairs_checked += 1;	    
+            if (dist_sq > max_dist_sq) {
+                max_dist_sq = dist_sq;
+                extreme[0] = i;
+                extreme[1] = j;
+            }
+        }
+    }
+
+    // stop the timer
+    end_time = MPI_Wtime();
+
+    // output the results
+    printf ("rank %d: elapsed time = %.4f seconds\n",rank,end_time-start_time);
+    printf ("rank %d: pairs checked = %d\n",rank,pairs_checked);    
+    printf ("rank %d: Extreme Distance = %.2f\n",rank,sqrt(max_dist_sq));
+    printf ("rank %d: Extreme Pair = %d %d\n",rank,extreme[0],extreme[1]);
+
+    // free memory allocated for dataset 
+    free(data);
+
+    MPI_Finalize();
+}

L19/mpi_extreme.sh

+#!/bin/bash
+#SBATCH -A cmda3634_rjh
+#SBATCH -p normal_q
+#SBATCH -t 5
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=4
+#SBATCH -o mpi_extreme.out
+
+# Go to the directory where the job was submitted
+cd $SLURM_SUBMIT_DIR
+
+# Load the modules we need for MPI
+module load matplotlib
+
+# Build the executable
+mpicc -o mpi_extreme mpi_extreme.c vec.c -lm
+
+# Run mpi_extreme
+printf 'Results with one rank:\n'
+mpiexec -n 1 --map-by ppr:$SLURM_NTASKS_PER_NODE:node ./mpi_extreme $1
+printf '\nResults with multiple ranks:\n'
+mpiexec -n $SLURM_NTASKS --map-by ppr:$SLURM_NTASKS_PER_NODE:node ./mpi_extreme $1
+

L19/mpi_extreme_approx.c

+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <mpi.h>
+#include "vec.h"
+
+int main (int argc, char* argv[]) {
+
+    MPI_Init (&argc, &argv);
+
+    // MPI_COMM_WORLD is the default communicator that contains all ranks
+    int rank, size;
+    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+    MPI_Comm_size(MPI_COMM_WORLD, &size);
+    char node_name[MPI_MAX_PROCESSOR_NAME];
+    int node_name_len;
+    MPI_Get_processor_name(node_name,&node_name_len);
+
+    // make sure we are not running on a login node!
+    if ((strcmp(node_name,"tinkercliffs1") == 0) || 
+            (strcmp(node_name,"tinkercliffs2") == 0)) {
+        printf ("error : running on login node %s!\n",node_name);
+        return 1;
+    }
+
+    // read the filename from the command line
+    if (argc < 4) {
+	printf ("command usage: %s %s %s %s\n",argv[0],"filename","num_pairs","seed");
+	return 1;
+    }
+    char* filename = argv[1];
+    int num_pairs = atoi(argv[2]);
+
+    // seed the random number generator using command line seed
+    srandom(atoi(argv[3]));
+
+    // open the text file for reading
+    FILE* fptr;
+    fptr = fopen(filename,"r");
+
+    // need to check for null
+    if (fptr == 0) {
+	printf ("Error opening data file %s.\n",filename);
+	exit(1);
+    }
+
+    // read the number of points and the dimension of each point
+    int num_points, dim;
+    if (fscanf(fptr,"%*c %d %d",&num_points, &dim) != 2) {
+        printf ("error reading the number of points and the dimension\n");
+        return 1;
+    }
+
+    // Read vectors from stdin and store them in a 2d array
+    double* data = (double*)malloc(num_points*dim*sizeof(double));
+    if (data == NULL) {
+        printf ("malloc return NULL pointer!\n");
+        return 1;
+    }
+    for (int i=0;i<num_points;i++) {
+        if (vec_read_file(fptr,data+i*dim,dim) != dim) {
+            printf ("error reading the next point from the file %s\n",filename);
+            return 1;
+        }
+    }
+
+    // close the data file
+    fclose(fptr);
+
+    // start the timer
+    double start_time, end_time;
+    start_time = MPI_Wtime();
+
+    // find the approximate extreme pair
+    double max_dist_sq = 0;
+    int pairs_checked = 0;
+    int extreme[2];
+    for (int p=0;p<num_pairs;p++) {
+	int i = random() % num_points;
+	int j = random() % num_points;
+	double dist_sq = vec_dist_sq(data+i*dim,data+j*dim,dim);
+	pairs_checked += 1;	    
+	if (dist_sq > max_dist_sq) {
+	    max_dist_sq = dist_sq;
+	    extreme[0] = i;
+	    extreme[1] = j;
+	}
+    }
+
+    // all nonzero ranks send their approximate extreme pair to rank 0
+    if (rank == 0) {
+	MPI_Status status;
+	int rank_extreme[2];
+	int rank_pairs_checked;
+	for (int src=1;src<size;src++) {
+	    MPI_Recv(&rank_pairs_checked,1,MPI_INT,src,0,MPI_COMM_WORLD,&status);
+	    pairs_checked += rank_pairs_checked;
+	    MPI_Recv(rank_extreme,2,MPI_INT,src,0,MPI_COMM_WORLD,&status);
+	    int i = rank_extreme[0];
+	    int j = rank_extreme[1];
+            double dist_sq = vec_dist_sq(data+i*dim,data+j*dim,dim);
+	    if (dist_sq > max_dist_sq) {
+		max_dist_sq = dist_sq;
+		extreme[0] = rank_extreme[0];
+		extreme[1] = rank_extreme[1];
+	    }
+	}
+    } else {
+	int dest = 0;
+	MPI_Send(&pairs_checked,1,MPI_INT,dest,0,MPI_COMM_WORLD);
+	MPI_Send(extreme,2,MPI_INT,dest,0,MPI_COMM_WORLD);
+    }
+
+    // stop the timer
+    end_time = MPI_Wtime();
+
+    // output the results
+    if (rank == 0) {
+	printf ("elapsed time = %.4f seconds\n",end_time-start_time);
+	printf ("pairs checked = %d\n",pairs_checked);
+	printf ("Approximate Extreme Distance = %.2f\n",sqrt(max_dist_sq));
+	printf ("Approximate Extreme Pair = %d %d\n",extreme[0],extreme[1]);
+    }
+
+    // free memory allocated for dataset 
+    free(data);
+
+    MPI_Finalize();
+}

L19/mpi_extreme_search.sh

+#!/bin/bash
+#SBATCH -A cmda3634_rjh
+#SBATCH -p normal_q
+#SBATCH -t 00:10:00
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=32
+#SBATCH -o mpi_extreme_search.out
+
+# Go to the directory where the job was submitted
+cd $SLURM_SUBMIT_DIR
+
+# Load the modules we need for MPI
+module load matplotlib
+
+# Build the executable
+mpicc -O3 -o mpi_extreme_approx mpi_extreme_approx.c vec.c -lm
+
+# run mpi_work on nodes*ntasks-per-node cores
+mpiexec -n $SLURM_NTASKS --map-by ppr:$SLURM_NTASKS_PER_NODE:node ./mpi_extreme_approx $1 $2 $3
+

L19/mpi_sum.c

+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <mpi.h>
+
+int main(int argc, char *argv[]) {
+
+    MPI_Init (&argc, &argv);
+
+    // MPI_COMM_WORLD is the default communicator that contains all ranks
+    int rank, size;
+    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+    MPI_Comm_size(MPI_COMM_WORLD, &size);
+    char node_name[MPI_MAX_PROCESSOR_NAME];
+    int node_name_len;
+    MPI_Get_processor_name(node_name,&node_name_len);
+
+    // make sure we are not running on a login node!
+    if ((strcmp(node_name,"tinkercliffs1") == 0) || 
+            (strcmp(node_name,"tinkercliffs2") == 0)) {
+        printf ("error : running on login node %s!\n",node_name);
+        return 1;
+    }
+
+    // get N from command line
+    if (argc < 2) {
+        printf ("Command usage : %s %s\n",argv[0],"N");
+        return 1;
+    }
+    long long N = atoll(argv[1]);
+
+    // start the timer
+    double start_time, end_time;
+    start_time = MPI_Wtime();
+
+    // calculate the sum
+    long long sum = 0;
+    for (long long i = 1+rank; i <= N;i+=size) {
+        sum += i;
+    }
+
+    // all nonzero ranks send their partial sums to rank 0
+    if (rank == 0) {
+        long long rank_sum;
+        MPI_Status status;
+        for (int src = 1;src < size;src++) {
+            MPI_Recv(&rank_sum,1,MPI_LONG_LONG,src,0,MPI_COMM_WORLD,&status);
+            sum += rank_sum;
+        }
+    } else {
+        int dest = 0;
+        MPI_Send(&sum,1,MPI_LONG_LONG,dest,0,MPI_COMM_WORLD);
+    }
+
+    // stop the timer
+    end_time = MPI_Wtime();
+
+    // print results
+    printf ("rank %d (of %d) sum = %lld, N*(N+1)/2 = %lld\n",rank,size,sum,(N/2)*(N+1));
+    if (rank == 0) {
+        printf ("elapsed time = %.4f seconds\n",end_time-start_time);
+    }
+
+    MPI_Finalize();
+}

L19/mpi_sum.sh

+#!/bin/bash
+#SBATCH -A cmda3634_rjh
+#SBATCH -p normal_q
+#SBATCH -t 5
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=4
+#SBATCH -o mpi_sum.out
+
+# Go to the directory where the job was submitted
+cd $SLURM_SUBMIT_DIR
+
+# Load the modules we need for MPI
+module load matplotlib
+
+# Build the executable
+mpicc -o mpi_sum mpi_sum.c vec.c -lm
+
+# Run mpi_sum
+printf 'Results with one rank:\n'
+mpiexec -n 1 --map-by ppr:$SLURM_NTASKS_PER_NODE:node ./mpi_sum $1
+printf '\nResults with multiple ranks:\n'
+mpiexec -n $SLURM_NTASKS --map-by ppr:$SLURM_NTASKS_PER_NODE:node ./mpi_sum $1
+

L19/vec.c

+#include <stdio.h>
+
+// v = 0
+void vec_zero (double v[], int dim) {
+    for (int i=0;i<dim;i++) {
+        v[i] = 0;
+    }
+}
+
+// read a vector from stdin
+// returns the number of elements read in
+int vec_read_stdin (double v[], int dim) {
+    for (int i=0;i<dim;i++) {
+        if (scanf("%lf",&(v[i])) != 1) { // could also use v+i
+            return i;
+        }
+    }
+    return dim;
+}
+
+// read a vector from a file
+// returns the number of elements read in
+int vec_read_file (FILE* fptr, double v[], int dim) {
+    for (int i=0;i<dim;i++) {
+        if (fscanf(fptr,"%lf",&(v[i])) != 1) { // could also use v+i
+            return i;
+        }
+    }
+    return dim;
+}
+
+// w = u + v
+void vec_add (double u[], double v[], double w[], int dim) {
+    for (int i=0;i<dim;i++) {
+        w[i] = u[i] + v[i];
+    }
+}
+
+// w = cv
+void vec_scalar_mult (double v[], double c, double w[], int dim) {
+    for (int i=0;i<dim;i++) {
+        w[i] = c*v[i];
+    }
+}
+
+// print a vector using the given format specifier
+void vec_print (double v[], char* format, int dim) {
+    for (int i=0;i<dim;i++) {
+        printf (format,v[i]);
+    }
+    printf ("\n");
+}
+
+// calculate the norm squared of a vector
+double vec_norm_sq (double v[], int dim) {
+    double norm_sq = 0;
+    for (int i=0;i<dim;i++) {
+        norm_sq += v[i]*v[i];
+    }
+    return norm_sq;
+}
+
+// calculate the distance squared between two vectors
+double vec_dist_sq (double u[], double v[], int dim) {
+    double dist_sq = 0;
+    for (int i=0;i<dim;i++) {
+        dist_sq += (u[i]-v[i])*(u[i]-v[i]);
+    }
+    return dist_sq;
+}
+
+// performs the copy v->data[i] = w->data[i] for all i
+void vec_copy (double v[], double w[], int dim) {
+    for (int i=0;i<dim;i++) {
+        v[i] = w[i];
+    }
+}

L19/vec.h

+#ifndef VEC_H
+#define VEC_H
+
+// v = 0
+void vec_zero (double v[], int dim);
+
+// read a vector from stdin
+// returns the number of elements read in
+int vec_read_stdin (double v[], int dim);
+
+// read a vector from a file
+// returns the number of elements read in
+int vec_read_file (FILE* fptr, double v[], int dim);
+
+// w = u + v
+void vec_add (double u[], double v[], double w[], int dim);
+
+// w = cv
+void vec_scalar_mult (double v[], double c, double w[], int dim);
+
+// print a vector using the given format specifier
+void vec_print (double v[], char* format, int dim);
+
+// calculate the norm squared of a vector
+double vec_norm_sq (double v[], int dim);
+
+// calculate the distance squared between two vectors
+double vec_dist_sq (double u[], double v[], int dim);
+
+// performs the copy v->data[i] = w->data[i] for all i
+void vec_copy (double v[], double w[], int dim);
+
+#endif