parallel-kmeans-cuda/kmeans.cpp at main · lebinary/parallel-kmeans-cuda · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
#include <unistd.h>
#include "kmeans.h"
#include "io.h"

void print_usage(char *prog_name) {
    fprintf(stderr, "Usage: %s -k <clusters> -d <dims> -i <input_file> -m <max_iter> -t <threshold> [-c] -s <seed> -v <version>\n", prog_name);
    fprintf(stderr, "  -k: number of clusters\n");
    fprintf(stderr, "  -d: dimension of points\n");
    fprintf(stderr, "  -i: input file name\n");
    fprintf(stderr, "  -m: maximum iterations\n");
    fprintf(stderr, "  -t: convergence threshold\n");
    fprintf(stderr, "  -c: output centroids (default: output assignments)\n");
    fprintf(stderr, "  -s: random seed\n");
    fprintf(stderr, "  -v: version (0=CPU, 1=CUDA Basic, 2=CUDA Shared Memory, 3=Thrust)\n");
}

void initialize_centers(double *points, double *centers, int n, int k, int d, int seed) {
    kmeans_srand(seed);

    int *selected = (int *)calloc(n, sizeof(int));

    for (int i = 0; i < k; i++) {
        int index;
        do {
            index = kmeans_rand() % n;
        } while (selected[index]);
        selected[index] = 1;

        for (int j = 0; j < d; j++) {
            centers[i * d + j] = points[index * d + j];
        }
    }

    free(selected);
}

int main(int argc, char *argv[]) {
    kmeans_args args = {0};
    args.max_iter = 150;
    args.threshold = 1e-5;
    args.version = 0;

    int opt;
    while ((opt = getopt(argc, argv, "k:d:i:m:t:cs:v:")) != -1) {
        switch (opt) {
            case 'k':
                args.k = atoi(optarg);
                break;
            case 'd':
                args.d = atoi(optarg);
                break;
            case 'i':
                args.input_file = optarg;
                break;
            case 'm':
                args.max_iter = atoi(optarg);
                break;
            case 't':
                args.threshold = atof(optarg);
                break;
            case 'c':
                args.output_centroids = true;
                break;
            case 's':
                args.seed = atoi(optarg);
                break;
            case 'v':
                args.version = atoi(optarg);
                break;
            default:
                print_usage(argv[0]);
                return 1;
        }
    }

    if (args.k <= 0 || args.d <= 0 || !args.input_file) {
        print_usage(argv[0]);
        return 1;
    }

    int n;
    double *points = read_points(args.input_file, &n, args.d);
    if (!points) {
        return 1;
    }
    args.n = n;

    double *centers = (double *)malloc(args.k * args.d * sizeof(double));
    int *assignments = (int *)malloc(n * sizeof(int));

    initialize_centers(points, centers, n, args.k, args.d, args.seed);

    kmeans_result result;
    switch(args.version) {
      case 0:
        result = kmeans_cpu(points, centers, assignments, &args);
        break;
      case 1:
        result = kmeans_kernel(points, centers, assignments, &args);
        break;
      case 2:
        result = kmeans_shared(points, centers, assignments, &args);
        break;
      case 3:
        result = kmeans_thrust(points, centers, assignments, &args);
        break;
      default:
        print_usage(argv[0]);
        return 1;
    }

    printf("%d,%lf\n", result.iterations, result.total_time / result.iterations);

    if (args.output_centroids) {
        write_centroids(centers, args.k, args.d);
    } else {
        write_assignments(assignments, n);
    }

    free_points(points);
    free(centers);
    free(assignments);

    return 0;
}